diff --git a/cecli/coders/__init__.py b/cecli/coders/__init__.py index 3fe9c0e3373..e5693180774 100644 --- a/cecli/coders/__init__.py +++ b/cecli/coders/__init__.py @@ -1,42 +1,57 @@ -from .agent_coder import AgentCoder -from .architect_coder import ArchitectCoder -from .ask_coder import AskCoder -from .base_coder import Coder -from .context_coder import ContextCoder -from .copypaste_coder import CopyPasteCoder -from .editblock_coder import EditBlockCoder -from .editblock_fenced_coder import EditBlockFencedCoder -from .editor_diff_fenced_coder import EditorDiffFencedCoder -from .editor_editblock_coder import EditorEditBlockCoder -from .editor_whole_coder import EditorWholeFileCoder -from .hashline_coder import HashLineCoder -from .help_coder import HelpCoder -from .patch_coder import PatchCoder -from .sub_agent_coder import SubAgentCoder -from .udiff_coder import UnifiedDiffCoder -from .udiff_simple import UnifiedDiffSimpleCoder -from .wholefile_coder import WholeFileCoder - -# from .single_wholefile_func_coder import SingleWholeFileFunctionCoder +"""Coder module with lazy imports to reduce startup memory.""" __all__ = [ - HelpCoder, - AskCoder, - Coder, - EditBlockCoder, - EditBlockFencedCoder, - WholeFileCoder, - PatchCoder, - UnifiedDiffCoder, - UnifiedDiffSimpleCoder, - # SingleWholeFileFunctionCoder, - ArchitectCoder, - EditorEditBlockCoder, - EditorWholeFileCoder, - EditorDiffFencedCoder, - ContextCoder, - AgentCoder, - CopyPasteCoder, - HashLineCoder, - SubAgentCoder, + "HelpCoder", + "AskCoder", + "Coder", + "EditBlockCoder", + "EditBlockFencedCoder", + "WholeFileCoder", + "PatchCoder", + "UnifiedDiffCoder", + "UnifiedDiffSimpleCoder", + "ArchitectCoder", + "EditorEditBlockCoder", + "EditorWholeFileCoder", + "EditorDiffFencedCoder", + "ContextCoder", + "AgentCoder", + "CopyPasteCoder", + "HashLineCoder", + "SubAgentCoder", ] + +# Module name mapping (snake_case to class name) +_MODULE_MAP = { + "HelpCoder": ".help_coder", + "AskCoder": ".ask_coder", + "Coder": ".base_coder", + "EditBlockCoder": ".editblock_coder", + "EditBlockFencedCoder": ".editblock_fenced_coder", + "WholeFileCoder": ".wholefile_coder", + "PatchCoder": ".patch_coder", + "UnifiedDiffCoder": ".udiff_coder", + "UnifiedDiffSimpleCoder": ".udiff_simple", + "ArchitectCoder": ".architect_coder", + "EditorEditBlockCoder": ".editor_editblock_coder", + "EditorWholeFileCoder": ".editor_whole_coder", + "EditorDiffFencedCoder": ".editor_diff_fenced_coder", + "ContextCoder": ".context_coder", + "AgentCoder": ".agent_coder", + "CopyPasteCoder": ".copypaste_coder", + "HashLineCoder": ".hashline_coder", + "SubAgentCoder": ".sub_agent_coder", +} + + +def __getattr__(name): + if name in _MODULE_MAP: + import importlib + + mod = importlib.import_module(_MODULE_MAP[name], __package__) + return getattr(mod, name) + raise AttributeError(f"module {__name__!r} has no attribute {name!r}") + + +def __dir__(): + return __all__ diff --git a/cecli/coders/agent_coder.py b/cecli/coders/agent_coder.py index 2f558dcb8fe..94d61738442 100644 --- a/cecli/coders/agent_coder.py +++ b/cecli/coders/agent_coder.py @@ -194,6 +194,7 @@ def _get_agent_config(self): "todo_list", "sub_agents", "skills", + "servers", }, ) ) @@ -357,6 +358,7 @@ def _calculate_context_block_tokens(self, force=False): "git_status", "symbol_outline", "skills", + "servers", "sub_agents", "loaded_skills", ] @@ -391,6 +393,8 @@ def _generate_context_block(self, block_name): content = self.get_todo_list() elif block_name == "skills": content = self.get_skills_context() + elif block_name == "servers": + content = self.get_servers_context() elif block_name == "loaded_skills": content = self.get_skills_content() elif block_name == "sub_agents" and ( @@ -624,8 +628,8 @@ def get_context_summary(self): result += f" ({percentage:.1f}% of limit)" if percentage > 80: result += "\n\n⚠ **Context is getting full!**\n" - result += "- Remove non-essential files via the `ContextManager` tool.\n" - result += "- Keep only essential files in context for best performance" + result += "- Remove non-essential files, skills and tool servers via the `ResourceManager` tool.\n" + result += "- Keep only essential files, skills, and MCP servers in context for best performance" result += "\n" if not hasattr(self, "context_blocks_cache"): self.context_blocks_cache = {} @@ -659,7 +663,9 @@ def get_environment_info(self): result += f"- Git repository: {rel_repo_dir} with {num_files:,} files\n" except Exception: result += "- Git repository: active but details unavailable\n" - else: + if self.mcp_manager and self.mcp_manager.connected_servers: + num_mcp_servers = len(self.mcp_manager.connected_servers) + result += f"- Connected MCP servers: {num_mcp_servers}\n" result += "- Git repository: none\n" result += "" return result @@ -806,9 +812,7 @@ async def gather_and_await(): if self.auto_lint and used_write_tool: edited = list(self.files_edited_by_tools) lint_coro = self.lint_edited(edited, show_output=False) - lint_errors, interrupted = await self.coroutines.interruptible( - lint_coro, self.interrupt_event - ) + lint_errors, interrupted = await interruptible(lint_coro, self.interrupt_event) if interrupted: raise KeyboardInterrupt("Interrupted during linting") @@ -928,9 +932,7 @@ async def reply_completed(self): ) self.io.tool_output(waiting_msg) sleep_coro = asyncio.sleep(command_timeout / 2) - _res, interrupted = await self.coroutines.interruptible( - sleep_coro, self.interrupt_event - ) + _res, interrupted = await interruptible(sleep_coro, self.interrupt_event) if interrupted: raise KeyboardInterrupt("Interrupted while waiting for background commands") return True @@ -1329,7 +1331,7 @@ async def check_for_file_mentions(self, content): Override parent's method to disable implicit file mention handling in agent mode. Files should only be added via explicit tool commands - (`ContextManager`). + (`ResourceManager`). """ pass @@ -1483,6 +1485,81 @@ def get_skills_content(self): self.io.tool_error(f"Error generating skills content context: {str(e)}") return None + def get_servers_context(self): + """ + Generate a context block for available MCP servers. + + Categorizes servers as: + - Active: Connected and passing includelist/excludelist filters + - Inactive: Connected but filtered out by includelist/excludelist + - Available (Disconnected): Managed but not currently connected + + Returns: + Formatted context block string or None if no servers available + """ + if not self.use_enhanced_context: + return None + try: + if not self.mcp_manager: + return None + + all_servers = self.mcp_manager.servers + connected_servers = self.mcp_manager.connected_servers + connected_server_names = {s.name for s in connected_servers} + + if not all_servers: + return None + + # Apply registered_servers filtering to determine active vs inactive + incl = self.registered_servers.get("included", set()) + excl = self.registered_servers.get("excluded", set()) + + active_servers = [] + inactive_servers = [] + for server in connected_servers: + name = server.name + if incl and name not in incl: + inactive_servers.append(name) + elif name in excl: + inactive_servers.append(name) + else: + active_servers.append(name) + + # Servers managed but not currently connected + disconnected_servers = [ + server.name for server in all_servers if server.name not in connected_server_names + ] + + result = '\n' + result += "## Connected MCP Servers\n\n" + + if active_servers: + result += f"Active ({len(active_servers)}):\n" + for name in sorted(active_servers): + result += f"- {name}\n" + result += "\n" + + if inactive_servers: + result += f"Inactive (Filtered) ({len(inactive_servers)}):\n" + for name in sorted(inactive_servers): + result += f"- {name}\n" + result += "\n" + + if disconnected_servers: + result += f"Available (Disconnected) ({len(disconnected_servers)}):\n" + for name in sorted(disconnected_servers): + result += f"- {name}\n" + result += "\n" + + if not active_servers and not inactive_servers and not disconnected_servers: + result += "No MCP servers currently available.\n\n" + + result += "" + return result + except Exception as e: + self.io.tool_error(f"Error generating servers context: {str(e)}") + return None + def get_sub_agents_context(self): """ Generate a context block for registered sub-agents. diff --git a/cecli/coders/architect_coder.py b/cecli/coders/architect_coder.py index 9eb6e53d68e..99d1ddff92e 100644 --- a/cecli/coders/architect_coder.py +++ b/cecli/coders/architect_coder.py @@ -45,7 +45,6 @@ async def reply_completed(self): kwargs["args"] = self.args kwargs["suggest_shell_commands"] = False kwargs["map_tokens"] = 0 - kwargs["total_cost"] = self.total_cost kwargs["cache_prompts"] = False kwargs["num_cache_warming_pings"] = 0 kwargs["summarize_from_coder"] = False diff --git a/cecli/coders/base_coder.py b/cecli/coders/base_coder.py index 1738092fb6e..ceeb4862a3d 100755 --- a/cecli/coders/base_coder.py +++ b/cecli/coders/base_coder.py @@ -31,12 +31,6 @@ from urllib.parse import urlparse from uuid import uuid4 as generate_unique_id -import httpx -from litellm import experimental_mcp_client -from litellm.types.utils import ModelResponse -from prompt_toolkit.patch_stdout import patch_stdout -from rich.console import Console - import cecli.prompts.utils.system as prompts from cecli import __version__, models, urls, utils from cecli.commands import Commands, SwitchCoderSignal @@ -111,6 +105,28 @@ def wrap_fence(name): wrap_fence("sourcecode"), ] +# Map of edit_format values to coder class names. +# Used by Coder.create() to find the right coder class by edit_format +# without importing all coder modules. +EDIT_FORMAT_MAP = { + "help": "HelpCoder", + "ask": "AskCoder", + "diff": "EditBlockCoder", + "diff-fenced": "EditBlockFencedCoder", + "whole": "WholeFileCoder", + "patch": "PatchCoder", + "udiff": "UnifiedDiffCoder", + "udiff-simple": "UnifiedDiffSimpleCoder", + "architect": "ArchitectCoder", + "editor-diff": "EditorEditBlockCoder", + "editor-whole": "EditorWholeFileCoder", + "editor-diff-fenced": "EditorDiffFencedCoder", + "context": "ContextCoder", + "agent": "AgentCoder", + "hashline": "HashLineCoder", + "subagent": "SubAgentCoder", +} + class UsageMeta(type): """Metaclass that provides shared accumulator properties across all Coder subclasses. @@ -335,9 +351,10 @@ async def create( res = coders.CopyPasteCoder(main_model, io, args=args, **kwargs) if not res: - for coder in coders.__all__: - if hasattr(coder, "edit_format") and coder.edit_format == edit_format: - res = coder(main_model, io, args=args, **kwargs) + coder_name = EDIT_FORMAT_MAP.get(edit_format) + if coder_name: + coder_cls = getattr(coders, coder_name) + res = coder_cls(main_model, io, args=args, **kwargs) if res is not None: if from_coder: @@ -359,11 +376,7 @@ async def create( res.original_kwargs = dict(kwargs) return res - valid_formats = [ - str(c.edit_format) - for c in coders.__all__ - if hasattr(c, "edit_format") and c.edit_format is not None - ] + valid_formats = list(EDIT_FORMAT_MAP.keys()) raise UnknownEditFormat(edit_format, valid_formats) async def clone(self, **kwargs): @@ -1470,6 +1483,8 @@ async def run(self, with_message=None, preproc=True): return await self._run_linear(with_message, preproc) if self.io.prompt_session: + from prompt_toolkit.patch_stdout import patch_stdout + with patch_stdout(raw=True): return await self._run_parallel(with_message, preproc) else: @@ -1954,7 +1969,10 @@ async def check_for_urls(self, inp: str) -> List[str]: def keyboard_interrupt(self): # Ensure cursor is visible on exit - Console().show_cursor(True) + if not self.tui: + from rich.console import Console + + Console().show_cursor(True) self.io.tool_warning("^C KeyboardInterrupt") self.interrupt_event.set() @@ -1973,13 +1991,25 @@ async def compact_context_if_needed(self, force=False, message=""): done_messages = manager.get_messages_dict(MessageTag.DONE) cur_messages = manager.get_messages_dict(MessageTag.CUR) diff_messages = manager.get_messages_dict(MessageTag.DIFFS) + all_messages = manager.get_messages_dict() # Exclude first cur_message since that's the user's initial input done_tokens = self.summarizer.count_tokens(done_messages) cur_tokens = self.summarizer.count_tokens(cur_messages[1:] if len(cur_messages) > 1 else []) diff_tokens = self.summarizer.count_tokens(diff_messages) + all_tokens = self.summarizer.count_tokens(all_messages) + combined_tokens = done_tokens + cur_tokens + diff_tokens + if force or ( + all_tokens >= self.context_compaction_max_tokens * 0.9 + and ConversationService.get_chunks(self).last_clear_count > 10 + ): + manager.clear_tag(MessageTag.DIFFS) + manager.clear_tag(MessageTag.FILE_CONTEXTS) + ConversationService.get_files(self).clear_file_cache() + ConversationService.get_chunks(self).flush_removals() + if not force and combined_tokens < self.context_compaction_max_tokens: return @@ -2881,6 +2911,8 @@ async def _execute_local_tools(self, tool_calls): async def _execute_mcp_tools(self, server, tool_calls): """Execute MCP tools via LiteLLM.""" + import httpx + tool_responses = [] try: # Connect to the server once @@ -2928,6 +2960,8 @@ async def _execute_mcp_tools(self, server, tool_calls): continue async def do_tool_call(): + from litellm import experimental_mcp_client + return await experimental_mcp_client.call_openai_tool( session=session, openai_tool=new_tool_call, @@ -3366,6 +3400,8 @@ async def check_for_file_mentions(self, content): return prompts.added_files.format(fnames=", ".join(added_fnames)) async def send(self, messages, model=None, functions=None, tools=None): + from litellm.types.utils import ModelResponse + self.interrupt_event.clear() self.got_reasoning_content = False self.ended_reasoning_content = False @@ -3452,6 +3488,8 @@ async def send(self, messages, model=None, functions=None, tools=None): self.io.ai_output(json.dumps(args, indent=4)) async def show_send_output(self, completion): + from litellm.types.utils import ModelResponse + if self.verbose: print(completion) diff --git a/cecli/coders/copypaste_coder.py b/cecli/coders/copypaste_coder.py index 65dd32d5a1f..ec92f522e4b 100644 --- a/cecli/coders/copypaste_coder.py +++ b/cecli/coders/copypaste_coder.py @@ -78,14 +78,13 @@ def _init_prompts_from_selected_edit_format(self): coders = None target_coder_class = None - if coders is not None: - for coder_cls in getattr(coders, "__all__", []): - if ( - hasattr(coder_cls, "edit_format") - and coder_cls.edit_format == selected_edit_format - ): - target_coder_class = coder_cls - break + if coders is not None and selected_edit_format: + # Use EDIT_FORMAT_MAP from base_coder to find the right coder class by edit_format + from cecli.coders.base_coder import EDIT_FORMAT_MAP + + coder_name = EDIT_FORMAT_MAP.get(selected_edit_format) + if coder_name: + target_coder_class = getattr(coders, coder_name, None) # Mirror prompt pack + edit_format where available. if target_coder_class is not None: diff --git a/cecli/commands/__init__.py b/cecli/commands/__init__.py index ae4e83a84cd..8095f367e9e 100644 --- a/cecli/commands/__init__.py +++ b/cecli/commands/__init__.py @@ -35,6 +35,7 @@ from .hot_reload import HotReloadCommand from .include_skill import IncludeSkillCommand from .lint import LintCommand +from .list_mcp import ListMcpCommand from .list_sessions import ListSessionsCommand from .list_skills import ListSkillsCommand from .load import LoadCommand @@ -123,6 +124,7 @@ CommandRegistry.register(SwitchAgentCommand) CommandRegistry.register(IncludeSkillCommand) CommandRegistry.register(LintCommand) +CommandRegistry.register(ListMcpCommand) CommandRegistry.register(ListSessionsCommand) CommandRegistry.register(ListSkillsCommand) CommandRegistry.register(LoadCommand) @@ -210,6 +212,7 @@ "LoadCommand", "LoadHookCommand", "LoadMcpCommand", + "ListMcpCommand", "LoadSessionCommand", "LoadSkillCommand", "LsCommand", diff --git a/cecli/commands/agent_model.py b/cecli/commands/agent_model.py index ceaa02ebf5f..758bb0da41e 100644 --- a/cecli/commands/agent_model.py +++ b/cecli/commands/agent_model.py @@ -47,7 +47,6 @@ async def execute(cls, io, coder, args, **kwargs): kwargs["main_model"] = model kwargs["edit_format"] = coder.edit_format # Keep the same edit format kwargs["suggest_shell_commands"] = False - kwargs["total_cost"] = coder.total_cost kwargs["num_cache_warming_pings"] = 0 kwargs["summarize_from_coder"] = False kwargs["done_messages"] = [] diff --git a/cecli/commands/editor_model.py b/cecli/commands/editor_model.py index b76190a56d5..28725a20466 100644 --- a/cecli/commands/editor_model.py +++ b/cecli/commands/editor_model.py @@ -46,7 +46,6 @@ async def execute(cls, io, coder, args, **kwargs): kwargs["main_model"] = model kwargs["edit_format"] = coder.edit_format # Keep the same edit format kwargs["suggest_shell_commands"] = False - kwargs["total_cost"] = coder.total_cost kwargs["num_cache_warming_pings"] = 0 kwargs["summarize_from_coder"] = False kwargs["done_messages"] = [] diff --git a/cecli/commands/list_mcp.py b/cecli/commands/list_mcp.py new file mode 100644 index 00000000000..0cf14c56446 --- /dev/null +++ b/cecli/commands/list_mcp.py @@ -0,0 +1,71 @@ +from cecli.commands.utils.base_command import BaseCommand +from cecli.commands.utils.helpers import format_command_result + + +class ListMcpCommand(BaseCommand): + NORM_NAME = "list-mcp" + DESCRIPTION = "List all loaded and configured MCP servers." + + @classmethod + async def execute(cls, io, coder, args, **kwargs): + """Execute the list-mcp command.""" + if not coder.mcp_manager: + return format_command_result(io, cls.NORM_NAME, "MCP manager is not configured.") + + all_servers = coder.mcp_manager.servers + connected_servers = coder.mcp_manager.connected_servers + + connected_server_names = {s.name for s in connected_servers} + + # Apply per-coder registered_servers filtering to determine active status + incl = coder.registered_servers["included"] + excl = coder.registered_servers["excluded"] + + active_servers = [] + inactive_servers = [] + + for server in connected_servers: + name = server.name + # Same filtering logic used in base_coder.get_tool_list() + if incl and name not in incl: + inactive_servers.append(name) + elif name in excl: + inactive_servers.append(name) + else: + active_servers.append(name) + + configured_servers = [ + server for server in all_servers if server.name not in connected_server_names + ] + + result = [] + if active_servers: + result.append("Active MCP Servers:") + for name in sorted(active_servers): + result.append(f"- {name}") + else: + result.append("No MCP servers are active for this coder.") + + if inactive_servers: + result.append("") + result.append("Inactive (Filtered) MCP Servers:") + for name in sorted(inactive_servers): + result.append(f"- {name}") + + result.append("") + if configured_servers: + result.append("Configured MCP Servers:") + for server in sorted(configured_servers, key=lambda s: s.name): + result.append(f"- {server.name}") + else: + result.append("No other MCP servers are configured.") + + return format_command_result(io, cls.NORM_NAME, "\n".join(result)) + + @classmethod + def get_help(cls) -> str: + """Get help text for the list-mcp command.""" + help_text = super().get_help() + help_text += "\nUsage:\n" + help_text += " /list-mcp # Lists MCP servers with coder-sensitive active/inactive/configured status\n" + return help_text diff --git a/cecli/commands/model.py b/cecli/commands/model.py index 83bc18e6e5f..f09c9fc64c2 100644 --- a/cecli/commands/model.py +++ b/cecli/commands/model.py @@ -50,7 +50,6 @@ async def execute(cls, io, coder, args, **kwargs): kwargs["main_model"] = model kwargs["edit_format"] = new_edit_format kwargs["suggest_shell_commands"] = False - kwargs["total_cost"] = coder.total_cost kwargs["num_cache_warming_pings"] = 0 kwargs["summarize_from_coder"] = False kwargs["done_messages"] = [] diff --git a/cecli/commands/remove_mcp.py b/cecli/commands/remove_mcp.py index 6a08ef33a5f..2b32307ecd1 100644 --- a/cecli/commands/remove_mcp.py +++ b/cecli/commands/remove_mcp.py @@ -10,7 +10,7 @@ class RemoveMcpCommand(BaseCommand): NORM_NAME = "remove-mcp" - DESCRIPTION = "Remove a MCP server by name, or use '*' to remove all" + DESCRIPTION = "Remove (unload) a MCP server by name, or use '*' to remove all" @classmethod async def execute(cls, io, coder, args, **kwargs): diff --git a/cecli/commands/weak_model.py b/cecli/commands/weak_model.py index acff8a48e30..12d79f6cffd 100644 --- a/cecli/commands/weak_model.py +++ b/cecli/commands/weak_model.py @@ -46,7 +46,6 @@ async def execute(cls, io, coder, args, **kwargs): kwargs["main_model"] = model kwargs["edit_format"] = coder.edit_format # Keep the same edit format kwargs["suggest_shell_commands"] = False - kwargs["total_cost"] = coder.total_cost kwargs["num_cache_warming_pings"] = 0 kwargs["summarize_from_coder"] = False kwargs["done_messages"] = [] diff --git a/cecli/helpers/conversation/integration.py b/cecli/helpers/conversation/integration.py index a245f50df64..a9381d4ecbb 100644 --- a/cecli/helpers/conversation/integration.py +++ b/cecli/helpers/conversation/integration.py @@ -18,7 +18,7 @@ class ConversationChunks: def __init__(self, coder): self.coder = weakref.ref(coder) self.uuid = coder.uuid - self._last_clear_count = 0 + self.last_clear_count = 0 self._deferred_removals = set() @classmethod @@ -282,14 +282,14 @@ def cleanup_files(self) -> None: if diff_count > 0 and other_count > 0 and diff_count / other_count > 20: should_clear = True - self._last_clear_count += 1 + self.last_clear_count += 1 if ( should_clear - and self._last_clear_count >= 20 + and self.last_clear_count >= 20 and diff_tokens + other_tokens > coder.context_compaction_max_tokens * 0.5 ): - self._last_clear_count = 0 + self.last_clear_count = 0 # Clear all diff messages ConversationService.get_manager(coder).clear_tag(MessageTag.DIFFS) @@ -890,7 +890,7 @@ def add_static_context_blocks(self) -> None: """ Add static context blocks to conversation (priority 50). - Static blocks include: environment_info, directory_structure, skills + Static blocks include: environment_info, directory_structure, skills, servers, sub_agents """ coder = self.get_coder() if not coder: @@ -922,6 +922,10 @@ def add_static_context_blocks(self) -> None: block = coder._generate_context_block("skills") if block: message_blocks["skills"] = block + if "servers" in coder.allowed_context_blocks: + block = coder._generate_context_block("servers") + if block: + message_blocks["servers"] = block # Add static blocks to conversation manager with stable hash keys for block_type, block_content in message_blocks.items(): diff --git a/cecli/helpers/file_system/service.py b/cecli/helpers/file_system/service.py index 42ed8ff83ca..478f683ad47 100644 --- a/cecli/helpers/file_system/service.py +++ b/cecli/helpers/file_system/service.py @@ -71,16 +71,19 @@ def reset_instance(cls) -> None: cls._instance = None @classmethod - def get_instance(cls, root: str = ".", repo=None) -> "FileSystemService": + def get_instance(cls, root: str | None = None, repo=None) -> "FileSystemService": """ Return the global singleton. On first call, creates and builds the instance using root/repo. - Subsequent calls return the existing instance (root/repo - parameters are ignored). This ensures all agents share one - file index regardless of when they're spawned. + Subsequent calls return the existing instance. If a non-None root + is explicitly provided and differs from the current root, the + singleton is rebuilt (e.g. when a new coder is created in a + different working directory). """ if cls._instance is None: + cls._instance = cls._create(root=root or ".", repo=repo) + elif root is not None and root != cls._instance.root: cls._instance = cls._create(root=root, repo=repo) return cls._instance diff --git a/cecli/helpers/hashline.py b/cecli/helpers/hashline.py index 4d985578b20..c6904ef3d23 100644 --- a/cecli/helpers/hashline.py +++ b/cecli/helpers/hashline.py @@ -1,8 +1,6 @@ import difflib import re -from cecli.helpers.grep_ast.parsers import filename_to_lang -from cecli.helpers.grep_ast.tsl import get_language, get_parser from cecli.helpers.hashpos.hashpos import HashPos HASHLINE_PREFIX_RE = HashPos.HASH_PREFIX_RE @@ -41,7 +39,7 @@ def strip_hashline(text: str) -> str: def normalize_hashline(hashline_str: str) -> str: """ - Normalize a hashline string to the 4-character hash fragment. + Normalize a hashline string to the content id hash fragment. """ if hashline_str in ("@000", "000@"): return hashline_str @@ -430,8 +428,8 @@ def get_hashline_diff( original_lines = original_content.splitlines() text_lines = text.splitlines() # Get up to 3 lines of context before (ending at found_end) and after the insertion point - ctx_before = original_lines[max(0, found_end - 2) : found_end + 1] - ctx_after = original_lines[found_end + 1 : min(len(original_lines), found_end + 4)] + ctx_before = original_lines[max(0, found_end - 6) : found_end] + ctx_after = original_lines[found_end + 1 : min(len(original_lines), found_end + 7)] # Build a mini document with context so HashPos computes correct neighborhood hashes mini_lines = ctx_before + text_lines + ctx_after mini_text = "\n".join(mini_lines) @@ -453,8 +451,8 @@ def get_hashline_diff( original_lines = original_content.splitlines() text_lines = text.splitlines() # Get up to 3 lines of context before and after the range - ctx_before = original_lines[max(0, found_start - 3) : found_start] - ctx_after = original_lines[found_end + 1 : min(len(original_lines), found_end + 4)] + ctx_before = original_lines[max(0, found_start - 6) : found_start] + ctx_after = original_lines[found_end + 1 : min(len(original_lines), found_end + 7)] # Build a mini document with context so HashPos computes correct neighborhood hashes mini_lines = ctx_before + text_lines + ctx_after mini_text = "\n".join(mini_lines) @@ -479,8 +477,8 @@ def get_hashline_diff( # Strip line endings for difflib comparison but keep them in the actual lines diff = difflib.unified_diff( - [line.rstrip("\r\n") for line in find_lines], - [line.rstrip("\r\n") for line in replace_lines], + [strip_hashline(line.rstrip("\r\n")) for line in find_lines], + [strip_hashline(line.rstrip("\r\n")) for line in replace_lines], lineterm="", n=1, ) @@ -1450,6 +1448,9 @@ def get_indentation(line: str) -> int: return resolved_ops # Determine language from file path + from cecli.helpers.grep_ast.parsers import filename_to_lang + from cecli.helpers.grep_ast.tsl import get_language, get_parser + lang = filename_to_lang(file_path) if not lang: return resolved_ops @@ -1463,7 +1464,6 @@ def get_indentation(line: str) -> int: return resolved_ops source_lines = original_content.splitlines() - MAX_STEPS = 3 # Maximum expansion steps for resolved in resolved_ops: op = resolved["op"] @@ -1489,47 +1489,35 @@ def get_indentation(line: str) -> int: llm_end = min(llm_end, len(source_lines) - 1) # --- THE HEALING LOOP --- - # Try original bounds first (distance 0), then progressively expand - # outward in rounds. At each round d>=1, test exactly 4 candidates: - # 1. Both indices down by d lines (range shifts down) - # 2. Both indices up by d lines (range shifts up) - # 3. Start index down by d lines, end unchanged (partial expansion) - # 4. End index down by d lines, start unchanged (partial expansion) - # - # If multiple candidates are valid at a round, select using: - # 1. Longest resulting source code (preserve more code) - # 2. Partial expansions over full range shifts - # 3. Downward changes over upward changes - - found_valid = False - EFFECTIVE_STEPS = MAX_STEPS - - if llm_start == llm_end: - EFFECTIVE_STEPS = 1 - - for distance in range(EFFECTIVE_STEPS + 1): - if distance == 0: - round_candidates = [(0, 0)] - else: - round_candidates = [ - (-distance, 0), # Start down only (partial) - (+distance, 0), # Start up only (partial) - (0, +distance), # End down only (partial) - (0, -distance), # End up only (partial) - (-distance, +distance), # Both indices down - (+distance, -distance), # Both indices up - (+distance, +distance), # Expand outward (start up, end down) - (-distance, -distance), # Contract inward (start down, end up) - ] - - valid_at_round = [] - for start_shift, end_shift in round_candidates: + # Generate all combinations of offsets in {-2, -1, 0, 1, 2} for both + # start and end, then filter to syntactically valid edits and rank + # by cumulative movement (abs(start_shift) + abs(end_shift)) as a + # primary criterion to bias toward minimal range perturbation. + + offsets = [0, 1, -1, 2, -2] + all_candidates = [] + for start_shift in offsets: + for end_shift in offsets: candidate_start = max(0, llm_start - start_shift) candidate_end = min(len(source_lines) - 1, llm_end + end_shift) if candidate_end < candidate_start: continue + # Skip candidates that overlap with other operations' ranges + overlaps = False + for other_op in resolved_ops: + if other_op is resolved: + continue + other_start = other_op["start_idx"] + other_end = other_op["end_idx"] + if candidate_start <= other_end and candidate_end >= other_start: + overlaps = True + break + + if overlaps: + continue + # Skip candidates that would create duplicate adjacent content at edit boundaries if _would_create_duplicate_content( source_lines, candidate_start, candidate_end, repl_lines @@ -1549,6 +1537,9 @@ def get_indentation(line: str) -> int: is_downward = start_shift <= 0 # Negative/zero shift = moving down is_both = start_shift == end_shift # Whole range expansion/contraction closures = count_closures(test_source) + start_line_match = source_lines[candidate_start] == source_lines[llm_start] + end_line_match = source_lines[candidate_end] == source_lines[llm_end] + cumulative_movement = abs(start_shift) + abs(end_shift) # --- INDENTATION SCORING --- indent_score = 0 @@ -1575,45 +1566,48 @@ def get_indentation(line: str) -> int: ) == get_indentation(source_end_line): indent_score += 1 - valid_at_round.append( + all_candidates.append( { "start_idx": candidate_start, "end_idx": candidate_end, + "start_line_match": start_line_match, + "end_line_match": end_line_match, "source_len": len(test_source), "is_partial": is_partial, "is_downward": is_downward, "is_both": is_both, "closure_count": closures, "indent_score": indent_score, + "cumulative_movement": cumulative_movement, + "offsets": (start_shift, end_shift), } ) - if valid_at_round: - # Sort using the new hierarchy: - # 1. Fewest total closures (minimize structural pollution) - # 2. Indentation Score (Descending: 2 is better than 0) - # 3. Longest source (preserve more file content) - # 4. Partial expansions over full range shifts - # 5. Downward changes over upward changes - valid_at_round.sort( - key=lambda r: ( - -r["indent_score"], # Descending: larger score is better (2 > 1 > 0) - -r["source_len"], # Descending: larger is better - r["closure_count"], # Ascending: smaller is better - not r["is_partial"], # Booleans: False comes before True - not r["is_downward"], # Booleans: False comes before True - r["is_both"], # Booleans: False comes before True - ) + if all_candidates: + # Sort using the new hierarchy: + # 1. Cumulative movement (ascending — less movement is better) + # 2. Fewest total closures (minimize structural pollution) + # 3. Indentation Score (Descending: 2 is better than 0) + # 4. Longest source (preserve more file content) + # 5. Partial expansions over full range shifts + # 6. Downward changes over upward changes + all_candidates.sort( + key=lambda r: ( + not r["start_line_match"], # Descending: True first + not r["end_line_match"], # Descending: True first + r["cumulative_movement"], # Ascending: smaller is better + -r["indent_score"], # Descending: larger score is better (2 > 1 > 0) + -r["source_len"], # Descending: larger is better + r["closure_count"], # Ascending: smaller is better + not r["is_partial"], # Booleans: False comes before True + not r["is_downward"], # Booleans: False comes before True + r["is_both"], # Booleans: False comes before True ) + ) - best = valid_at_round[0] - resolved["start_idx"] = best["start_idx"] - resolved["end_idx"] = best["end_idx"] - found_valid = True - break - - if not found_valid: - pass + best = all_candidates[0] + resolved["start_idx"] = best["start_idx"] + resolved["end_idx"] = best["end_idx"] return resolved_ops @@ -1758,6 +1752,12 @@ def apply_hashline_operations( resolved_ops = _merged_contained_ranges(resolved_ops) # Merge contiguous replace operations resolved_ops = _merge_replace_operations(resolved_ops) + + # Sort by start_idx descending to apply from bottom to top + # When operations have same start_idx, apply in order: insert, replace, delete + # This ensures correct behavior when multiple operations target the same line + resolved_ops.sort(key=sort_ranges) + if file_path: # Apply tree-sitter based closure safeguard to snap boundaries to AST nodes resolved_ops = _apply_closure_safeguard(original_content, resolved_ops, file_path) @@ -1766,11 +1766,6 @@ def apply_hashline_operations( source_lines = original_content.splitlines() resolved_ops = _fix_duplicate_content_boundaries(source_lines, resolved_ops) - # Sort by start_idx descending to apply from bottom to top - # When operations have same start_idx, apply in order: insert, replace, delete - # This ensures correct behavior when multiple operations target the same line - resolved_ops.sort(key=sort_ranges) - successful_ops = [] # Loop to apply operations in sorted order (bottom-to-top) for resolved in resolved_ops: diff --git a/cecli/helpers/hashpos/hashpos.py b/cecli/helpers/hashpos/hashpos.py index 516052012c9..280e3394ec4 100644 --- a/cecli/helpers/hashpos/hashpos.py +++ b/cecli/helpers/hashpos/hashpos.py @@ -4,37 +4,67 @@ class HashPos: - B64 = "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789~_" - # Regex pattern for HashPos format: {4-char-hash}:: - HASH_PREFIX_RE = re.compile(r"^([0-9a-zA-Z\~_@]{4})::") - # Regex for normalization: 4 hash chars optionally followed by '::' - NORMALIZE_RE = re.compile(r"^([0-9a-zA-Z\~_@]{4})(?:)?::") - # Regex for a raw 4-character fragment - FRAGMENT_RE = re.compile(r"^[0-9a-zA-Z\~_@]{4}$") + B256 = ( + "ABCDEFGHIJKLMNOP" + "QRSTUVWXYZabcdef" + "ghijklmnopqrstuv" + "wxyz0123456789~_" + "áéíóúñüöäßåøæçèà" + "ùîôûбгджзийлпфцч" + "шщъыьэюя的是不了人我在有" + "他这为之大来以个中上们到说国和学" + "あいうえおかきくけこさしすせそた" + "ちつてとアイウエオカキクケコサシ" + "スセソタチツテトαβγδεζηθ" + "ικλμνξπ要会出就道也时年得" + "生自下而过能可对行没发用天作方成" + "者多日都三小机把理实心看起样好当" + "点本民事其然想经去种动全意面前所" + "业定现将法新明问度但最美月手走信" + ) + + # We use a regex-safe character class string for compiling patterns + _B256_REGEX_SET = ( + "A-Za-z0-9~_" + "áéíóúñüöäßåøæçèà" + "ùîôûбгджзийлпфцч" + "шщъыьэюя的是不了人我在有" + "他这为之大来以个中上们到说国和学" + "あいうえおかきくけこさしすせそた" + "ちつてとアイウエオカキクケコサシ" + "スセソタチツテトαβγδεζηθ" + "ικλμνξπ要会出就道也时年得" + "生自下而过能可对行没发用天作方成" + "者多日都三小机把理实心看起样好当" + "点本民事其然想经去种动全意面前所" + "业定现将法新明问度但最美月手走信" + ) + + # Regex pattern for HashPos format: {3-char-hash}:: + HASH_PREFIX_RE = re.compile(rf"^([{_B256_REGEX_SET}]{{3}})::") + # Regex for normalization: 3 hash chars optionally followed by '::' + NORMALIZE_RE = re.compile(rf"^([{_B256_REGEX_SET}]{{3}})(?:)?::") + # Regex for a raw 3-character fragment + FRAGMENT_RE = re.compile(rf"^[{_B256_REGEX_SET}]{{3}}$") def __init__(self, source_text: str = ""): self.lines = source_text.splitlines() self.total = len(self.lines) - def _get_region_bits(self, line_idx: int) -> tuple[int, int]: + def _get_region_val(self, line_idx: int) -> int: """ - Uses line_idx modulo 16 (4 bits) to get two 2-bit flags (b1, b2). - This guarantees up to 16 consecutive repeating lines get unique spatial anchors. + Uses line_idx modulo 16 (4 bits). + Guarantees up to 16 consecutive repeating lines get unique spatial anchors. """ - mod_val = line_idx % 16 - - # Split the 4-bit modulo value into two separate 2-bit flags - b1 = (mod_val >> 2) & 3 # Top 2 bits (mask with 0b11) - b2 = mod_val & 3 # Bottom 2 bits - return b1, b2 + return line_idx % 16 def _get_neighborhood_hash(self, line_idx: int) -> int: """ Creates a 20-bit digest using the current line and the 3 lines before and after it. """ - start = max(0, line_idx - 3) - end = min(self.total, line_idx + 4) + start = max(0, line_idx - 2) + end = min(self.total, line_idx + 3) context_window = "\n".join(self.lines[start:end]) full_hash = xxhash.xxh3_64_intdigest(context_window.encode("utf-8")) @@ -51,22 +81,20 @@ def generate_private_id(self, text: str) -> str: def generate_public_id(self, text: str, line_idx: int) -> str: """ - Generates a 4-char Base64 ID combining modulo buckets and context hash. - Layout: [2-bit b1] [2-bit b2] [10-bit Hash A] [10-bit Hash B] + Generates a 3-char Base256 ID combining a 4-bit modulo bucket and a 20-bit context hash. + Layout: [4-bit Region] [20-bit Neighborhood Hash] = 24 bits total. + Each Base256 char holds 8 bits (3 chars * 8 = 24 bits). """ - b1, b2 = self._get_region_bits(line_idx) + region_val = self._get_region_val(line_idx) neighborhood_hash = self._get_neighborhood_hash(line_idx) - # Split the 20-bit hash into two 10-bit halves - hash_a = (neighborhood_hash >> 10) & 0x3FF - hash_b = neighborhood_hash & 0x3FF + # Pack the 24-bit integer + packed = (region_val << 20) | neighborhood_hash - # Construct the mixed 24-bit integer - packed = (b1 << 22) | (b2 << 20) | (hash_a << 10) | hash_b res = "" - for _ in range(4): - res += self.B64[packed % 64] - packed //= 64 + for _ in range(3): + res += self.B256[packed % 256] + packed //= 256 return res def unpack_public_id(self, public_id: str) -> tuple[int, int]: @@ -75,16 +103,13 @@ def unpack_public_id(self, public_id: str) -> tuple[int, int]: """ packed = 0 for i, char in enumerate(public_id): - packed |= self.B64.index(char) << (6 * i) + packed |= self.B256.index(char) << (8 * i) - b1 = (packed >> 22) & 3 - b2 = (packed >> 20) & 3 - hash_a = (packed >> 10) & 0x3FF - hash_b = packed & 0x3FF - mod_val = (b1 << 2) | b2 - neighborhood_hash = (hash_a << 10) | hash_b + # Extract the 4-bit region (mask 0xF) and 20-bit hash (mask 0xFFFFF) + region_val = (packed >> 20) & 0xF + neighborhood_hash = packed & 0xFFFFF - return mod_val, neighborhood_hash + return region_val, neighborhood_hash def format_content(self, use_private_ids: bool = False, start_line: int = 1) -> str: formatted_lines = [] @@ -128,11 +153,6 @@ def modulo_distance(idx: int) -> int: def resolve_range(self, start_id: str, end_id: str) -> tuple[int, int]: """ Resolves a block range from two Public IDs. - - Logic: - 1. Resolve all candidates for both IDs (sorted by best match). - 2. Find the pair of (start, end) that are logically ordered. - 3. Returns (start_index, end_index) """ starts = self.resolve_to_lines(start_id) ends = self.resolve_to_lines(end_id) @@ -146,28 +166,17 @@ def resolve_range(self, start_id: str, end_id: str) -> tuple[int, int]: return s, e raise ValueError( - f"Found matches for {start_id} and {end_id}, but no logically ordered range or unique" - " matches." + f"Found matches for {start_id} and {end_id}, but no logically ordered range or unique matches." ) @staticmethod def strip_prefix(text: str) -> str: - r""" + """ Remove HashPos prefixes from the start of every line. - - Removes prefixes that match the pattern: "{4-char-hash}" - where the hash is exactly 4 characters from the set [0-9a-zA-Z\~_@] followed by '::'. - - Args: - text: Input text with HashPos prefixes - - Returns: - String with HashPos prefixes removed from each line """ lines = text.splitlines(keepends=True) result_lines = [] for line in lines: - # Remove the HashPos prefix if present stripped_line = HashPos.HASH_PREFIX_RE.sub("", line, count=1) result_lines.append(stripped_line) @@ -177,12 +186,6 @@ def strip_prefix(text: str) -> str: def extract_prefix(line: str) -> str: """ Extract the hash prefix from a line if it has a HashPos prefix. - - Args: - line: A line of text that may contain a HashPos prefix - - Returns: - The hash prefix (4 characters) if found, otherwise empty string """ match = HashPos.HASH_PREFIX_RE.match(line) if match: @@ -192,25 +195,11 @@ def extract_prefix(line: str) -> str: @staticmethod def normalize(hashpos_str: str) -> str: """ - Normalize a HashPos string to the 4-character hash fragment. - - Accepts HashPos strings in "{hash_prefix}::" format or a raw "{hash_prefix}" fragment. - Also extracts HashPos from strings that contain content after the HashPos, - e.g., "H7M5::Line 1" - - Args: - hashpos_str: HashPos string in various formats - - Returns: - str: The 4-character hash fragment - - Raises: - ValueError: If format is invalid + Normalize a HashPos string to the 3-character hash fragment. """ if hashpos_str is None: raise ValueError("HashPos string cannot be None") - # Check if it's already a raw fragment if HashPos.FRAGMENT_RE.match(hashpos_str): return hashpos_str @@ -218,9 +207,7 @@ def normalize(hashpos_str: str) -> str: if match: return match.group(1) - # If no pattern matches, raise error raise ValueError( f"Invalid HashPos format '{hashpos_str}'. " - r"Expected \"{content ID}\" " - r"where content ID is exactly 4 characters from the set [0-9a-zA-Z\~_@]." + r"Expected a 3-character string from the Base256 character set." ) diff --git a/cecli/helpers/responses.py b/cecli/helpers/responses.py index 0bc9cecf909..8a77db27bb4 100644 --- a/cecli/helpers/responses.py +++ b/cecli/helpers/responses.py @@ -1,14 +1,18 @@ +from __future__ import annotations + import json import re import time -from typing import List, Optional +from typing import TYPE_CHECKING, List, Optional import json_repair -from litellm.types.utils import ChatCompletionMessageToolCall, Function from cecli import utils from cecli.helpers import nested +if TYPE_CHECKING: + from litellm.types.utils import ChatCompletionMessageToolCall, Function # noqa + def preprocess_json(response: str) -> str: # This pattern matches any sequence of backslashes followed by @@ -35,6 +39,8 @@ def extract_tools_from_content_json(content: str) -> Optional[List[ChatCompletio Simple extraction of JSON-like structures that look like tool calls. This handles models that write JSON in text instead of using native calling. """ + from litellm.types.utils import ChatCompletionMessageToolCall, Function # noqa + if not content or ("{" not in content and "[" not in content): return None @@ -111,6 +117,8 @@ def extract_tools_from_content_xml(content: str) -> Optional[List[ChatCompletion """ + from litellm.types.utils import ChatCompletionMessageToolCall, Function # noqa + if not content or (" bool: Returns: Boolean indicating success or failure """ + from litellm import experimental_mcp_client + server = self.get_server(name) if not server: self._log_warning(f"MCP server not found: {name}") @@ -131,9 +132,7 @@ async def connect_server(self, name: str) -> bool: try: session = await server.connect() - tools = await litellm.experimental_mcp_client.load_mcp_tools( - session=session, format="openai" - ) + tools = await experimental_mcp_client.load_mcp_tools(session=session, format="openai") self._server_tools[server.name] = tools self._connected_servers.add(server) self._log_verbose(f"Connected to MCP server: {name}") diff --git a/cecli/models.py b/cecli/models.py index 190757a4dc4..3f593cb7b75 100644 --- a/cecli/models.py +++ b/cecli/models.py @@ -148,6 +148,7 @@ def __init__(self): self.cache_dir = handle_core_files(Path.home() / ".cecli" / "caches") self.cache_file = self.cache_dir / "model_prices_and_context_window.json" self.content = None + self._raw_content = None self.local_model_metadata = {} self.verify_ssl = True self._cache_loaded = False @@ -167,9 +168,9 @@ def _load_cache(self): cache_age = time.time() - self.cache_file.stat().st_mtime if cache_age < self.CACHE_TTL: try: - self.content = json.loads(self.cache_file.read_text()) + self._raw_content = self.cache_file.read_text() except json.JSONDecodeError: - self.content = None + self._raw_content = None except OSError: pass self._cache_loaded = True @@ -180,9 +181,13 @@ def _update_cache(self): response = requests.get(self.MODEL_INFO_URL, timeout=5, verify=self.verify_ssl) if response.status_code == 200: - self.content = response.json() + # Use json.dumps(response.json()) instead of response.text for + # compatibility with mocked responses in tests + parsed = response.json() + self._raw_content = json.dumps(parsed) try: - self.cache_file.write_text(json.dumps(self.content, indent=4)) + parsed = response.json() + self.cache_file.write_text(json.dumps(parsed, indent=4)) except OSError: pass except Exception as ex: @@ -192,21 +197,59 @@ def _update_cache(self): except OSError: pass + def _get_entry_from_raw(self, key): + """Parse a single model entry from raw JSON string without loading the entire dict.""" + if not self._raw_content: + return None + import re + + escaped_key = re.escape(key) + pattern = rf'(? 0: + ch = self._raw_content[pos] + if escape: + escape = False + elif ch == "\\": + escape = True + elif ch == '"': + in_string = not in_string + elif not in_string: + if ch == "{": + depth += 1 + elif ch == "}": + depth -= 1 + pos += 1 + if depth == 0: + entry_str = self._raw_content[start:pos] + return json.loads(entry_str) + return None + def get_model_from_cached_json_db(self, model): data = self.local_model_metadata.get(model) if data: return data self._load_cache() - if not self.content: + if not self._raw_content: self._update_cache() - if not self.content: + if not self._raw_content: return dict() - info = self.content.get(model, dict()) + info = self._get_entry_from_raw(model) if info: return info pieces = model.split("/") if len(pieces) == 2: - info = self.content.get(pieces[1]) + info = self._get_entry_from_raw(pieces[1]) if info and info.get("litellm_provider") == pieces[0]: return info return dict() diff --git a/cecli/prompts/agent.yml b/cecli/prompts/agent.yml index 4143d42d18e..ec1473e3f41 100644 --- a/cecli/prompts/agent.yml +++ b/cecli/prompts/agent.yml @@ -27,21 +27,22 @@ main_system: | ## FILE FORMAT File contents will be prefixed with identifiers. Each line starts with a case-sensitive content ID followed by `::`. These are used to target where editing tools will perform edits. - They are algorithmically generated, maintained, and subject to change. Do not search for these content IDs. Focus on the lines they identify. + They are generated and maintained by a custom algorithm and subject to change on edits. Do not search for these content IDs directly. You will not be able to generate them. + Focus on the lines they identify. **Example File** ``` - il9n::#!/usr/bin/env python3 - faoZ:: - uXdn::def example_method(): - WAR5:: return "example" - vwkS:: + 事tN::#!/usr/bin/env python3 + 看XX:: + 法然t::def example_method(): + ä个8:: return "example" + 都ъñ:: ``` ## Core Workflow 1. **Plan**: Start by using `UpdateTodoList` to outline the task. 2. **Explore**: Use discovery tools (`ExploreCode`, `Grep`, `Ls`) to research and gather understanding for you task. Modify search terms when errors are encountered. - 3. **Execute**: Mark files as editable with `ContextManager` before attempting edits. Proactively use skills if they are available. Review diff outputs after edit to ensure the proper changes were made. + 3. **Execute**: Mark files as editable with `ResourceManager` before attempting edits. Proactively use skills if they are available. Review diff outputs after edit to ensure the proper changes were made. 4. **Verify & Recover**: If an edit fails or introduces linting errors, fix the error immediately. Use `UndoChange` if the errors are too complex to incrementally modify. 5. **Yield**: Use the `Yield` tool after accomplishing the goal and verifying any changes made. Provide helpful summaries of any changes. @@ -51,7 +52,7 @@ main_system: | ## Operational Rules - **Scope**: No unrequested refactors. Avoid full-file rewrites. Only modify what you are asked to. - - **Hygiene**: Use `ContextManager`/`RemoveSkill` to evict unneeded files/skills immediately after use. + - **Hygiene**: Use `ResourceManager` to evict unneeded files/skills immediately after use. - **Outputs**: Tool calls trigger turns. Never include tool syntax in final user summaries. - **Sandbox**: Perform all verification and temp logic in `.cecli/temp`. - **Responses**: Reason out loud through the problem but be brief. @@ -70,7 +71,7 @@ system_reminder: | {lazy_prompt} {shell_cmd_reminder} - "" + try_again: | My previous exploration was insufficient. I will now adjust my strategy, use more specific search patterns, and manage my context more aggressively to find the correct solution. \ No newline at end of file diff --git a/cecli/prompts/subagent.yml b/cecli/prompts/subagent.yml index e10c1935e1d..6e3867a9852 100644 --- a/cecli/prompts/subagent.yml +++ b/cecli/prompts/subagent.yml @@ -12,21 +12,22 @@ main_system: | ## FILE FORMAT File contents will be prefixed with identifiers. Each line starts with a case-sensitive content ID followed by `::`. These are used to target where editing tools will perform edits. - They are algorithmically generated, maintained, and subject to change. Do not search for these content IDs. Focus on the lines they identify. + They are generated and maintained by a custom algorithm and subject to change on edits. Do not search for these content IDs directly. You will not be able to generate them. + Focus on the lines they identify. **Example File** ``` - il9n::#!/usr/bin/env python3 - faoZ:: - uXdn::def example_method(): - WAR5:: return "example" - vwkS:: + 事tN::#!/usr/bin/env python3 + 看XX:: + 法然t::def example_method(): + ä个8:: return "example" + 都ъñ:: ``` ## Core Workflow 1. **Plan**: Start by using `UpdateTodoList` to outline the task. 2. **Explore**: Use discovery tools (`ExploreCode`, `Grep`, `Ls`) to research and gather understanding for you task. Modify search terms when errors are encountered. - 3. **Execute**: Mark files as editable with `ContextManager` before attempting edits. Proactively use skills if they are available. Review diff outputs after edit to ensure the proper changes were made. + 3. **Execute**: Mark files as editable with `ResourceManager` before attempting edits. Proactively use skills if they are available. Review diff outputs after edit to ensure the proper changes were made. 4. **Verify & Recover**: If an edit fails or introduces linting errors, fix the error immediately. Use `UndoChange` if the errors are too complex to incrementally modify. 5. **Yield**: Use the `Yield` tool after accomplishing the goal and verifying any changes made. Provide helpful summaries of any changes. @@ -36,7 +37,7 @@ main_system: | ## Operational Rules - **Scope**: No unrequested refactors. Avoid full-file rewrites. Only modify what you are asked to. - - **Hygiene**: Use `ContextManager`/`RemoveSkill` to evict unneeded files/skills immediately after use. + - **Hygiene**: Use `ResourceManager` to evict unneeded files/skills immediately after use. - **Outputs**: Tool calls trigger turns. Never include tool syntax in final user summaries. - **Sandbox**: Perform all verification and temp logic in `.cecli/temp`. - **Responses**: Reason out loud through the problem but be brief. diff --git a/cecli/repomap.py b/cecli/repomap.py index 1f721ef0ca3..dbb9739d5c3 100644 --- a/cecli/repomap.py +++ b/cecli/repomap.py @@ -10,10 +10,7 @@ from importlib import resources from pathlib import Path -import tree_sitter from diskcache import Cache -from pygments.lexers import guess_lexer_for_filename -from pygments.token import Token from cecli.dump import dump from cecli.helpers.similarity import ( @@ -22,18 +19,9 @@ normalize_vector, ) from cecli.special import filter_important_files -from cecli.tools.utils.helpers import ToolError -# tree_sitter is throwing a FutureWarning warnings.simplefilter("ignore", category=FutureWarning) -from cecli.helpers.grep_ast import TreeContext, filename_to_lang # noqa: E402 -from cecli.helpers.grep_ast.tsl import ( # noqa: E402 - USING_TSL_PACK, - get_language, - get_parser, -) - # Define the Tag namedtuple with a default for specific_kind to maintain compatibility # with cached entries that might have been created with the old definition @@ -117,9 +105,11 @@ def __new__( SQLITE_ERRORS = (sqlite3.OperationalError, sqlite3.DatabaseError, OSError) -CACHE_VERSION = 7 -if USING_TSL_PACK: - CACHE_VERSION = 9 +# CACHE_VERSION determines tags cache format. +# Set to 9 to ensure fresh cache when tree-sitter features are available. +# tree_sitter is loaded lazily inside get_tags_raw(), so we don't +# import it at module level here. +CACHE_VERSION = 9 UPDATING_REPO_MAP_MESSAGE = "Updating repo map" @@ -517,6 +507,8 @@ def get_symbol_definition_location(self, file_path, symbol_name): Raises: ToolError: If the symbol is not found, not unique, or not a definition. """ + from cecli.tools.utils.helpers import ToolError + abs_path = self.io.root_abs_path(file_path) # Assuming io has this helper or similar rel_path = self.get_rel_fname(abs_path) # Ensure we use consistent relative path @@ -605,6 +597,13 @@ def check_import_match(self, definer, imports): return False def get_tags_raw(self, fname, rel_fname): + import tree_sitter + from pygments.lexers import guess_lexer_for_filename + from pygments.token import Token + + from cecli.helpers.grep_ast import filename_to_lang + from cecli.helpers.grep_ast.tsl import USING_TSL_PACK, get_language, get_parser + lang = filename_to_lang(fname) if not lang: return @@ -1279,6 +1278,8 @@ def get_ranked_tags_map_uncached( def render_tree( self, abs_fname, rel_fname, lois, line_numbers=False, start_line=None, end_line=None ): + from cecli.helpers.grep_ast import TreeContext + mtime = self.get_mtime(abs_fname) key = (rel_fname, tuple(sorted(lois)), mtime, start_line, end_line) @@ -1444,6 +1445,8 @@ def find_src_files(directory): def get_scm_fname(lang): + from cecli.helpers.grep_ast.tsl import USING_TSL_PACK + # Load the tags queries if USING_TSL_PACK: subdir = "tree-sitter-language-pack" diff --git a/cecli/resources/model-metadata.json b/cecli/resources/model-metadata.json index 8aee168a3e5..4af6e04ffc7 100644 --- a/cecli/resources/model-metadata.json +++ b/cecli/resources/model-metadata.json @@ -353,6 +353,39 @@ "cache_read_input_token_cost": 3e-7, "cache_creation_input_token_cost": 0.00000375 }, + "anthropic.claude-fable-5": { + "cache_creation_input_token_cost": 0.0000125, + "cache_creation_input_token_cost_above_1hr": 0.00002, + "cache_read_input_token_cost": 0.000001, + "input_cost_per_token": 0.00001, + "litellm_provider": "bedrock_converse", + "max_input_tokens": 1000000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "mode": "chat", + "output_cost_per_token": 0.00005, + "search_context_cost_per_query": { + "search_context_size_high": 0.01, + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01 + }, + "supports_adaptive_thinking": true, + "supports_assistant_prefill": false, + "supports_computer_use": true, + "supports_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_sampling_params": false, + "supports_tool_choice": true, + "supports_vision": true, + "supports_xhigh_reasoning_effort": true, + "supports_native_structured_output": true, + "supports_max_reasoning_effort": true, + "supports_output_config": true, + "bedrock_output_config_effort_ceiling": "xhigh" + }, "anthropic.claude-haiku-4-5-20251001-v1:0": { "cache_creation_input_token_cost": 0.00000125, "cache_creation_input_token_cost_above_1hr": 0.000002, @@ -557,6 +590,7 @@ "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, + "supports_sampling_params": false, "supports_tool_choice": true, "supports_vision": true, "supports_xhigh_reasoning_effort": true, @@ -581,6 +615,7 @@ "search_context_size_low": 0.01, "search_context_size_medium": 0.01 }, + "supports_adaptive_thinking": true, "supports_assistant_prefill": false, "supports_computer_use": true, "supports_function_calling": true, @@ -588,6 +623,7 @@ "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, + "supports_sampling_params": false, "supports_tool_choice": true, "supports_vision": true, "supports_xhigh_reasoning_effort": true, @@ -1019,6 +1055,7 @@ }, "au.anthropic.claude-haiku-4-5-20251001-v1:0": { "cache_creation_input_token_cost": 0.000001375, + "cache_creation_input_token_cost_above_1hr": 0.0000022, "cache_read_input_token_cost": 1.1e-7, "input_cost_per_token": 0.0000011, "litellm_provider": "bedrock_converse", @@ -1040,6 +1077,7 @@ }, "au.anthropic.claude-opus-4-6-v1": { "cache_creation_input_token_cost": 0.000006875, + "cache_creation_input_token_cost_above_1hr": 0.000011, "cache_read_input_token_cost": 5.5e-7, "input_cost_per_token": 0.0000055, "litellm_provider": "bedrock_converse", @@ -1069,6 +1107,7 @@ }, "au.anthropic.claude-opus-4-7": { "cache_creation_input_token_cost": 0.000006875, + "cache_creation_input_token_cost_above_1hr": 0.000011, "cache_read_input_token_cost": 5.5e-7, "input_cost_per_token": 0.0000055, "litellm_provider": "bedrock_converse", @@ -1089,6 +1128,7 @@ "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, + "supports_sampling_params": false, "supports_tool_choice": true, "supports_vision": true, "supports_xhigh_reasoning_effort": true, @@ -1113,6 +1153,7 @@ "search_context_size_low": 0.01, "search_context_size_medium": 0.01 }, + "supports_adaptive_thinking": true, "supports_assistant_prefill": false, "supports_computer_use": true, "supports_function_calling": true, @@ -1120,6 +1161,7 @@ "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, + "supports_sampling_params": false, "supports_tool_choice": true, "supports_vision": true, "supports_xhigh_reasoning_effort": true, @@ -1130,11 +1172,13 @@ }, "au.anthropic.claude-sonnet-4-5-20250929-v1:0": { "cache_creation_input_token_cost": 0.000004125, + "cache_creation_input_token_cost_above_1hr": 0.0000066, "cache_read_input_token_cost": 3.3e-7, "input_cost_per_token": 0.0000033, "input_cost_per_token_above_200k_tokens": 0.0000066, "output_cost_per_token_above_200k_tokens": 0.00002475, "cache_creation_input_token_cost_above_200k_tokens": 0.00000825, + "cache_creation_input_token_cost_above_1hr_above_200k_tokens": 0.0000132, "cache_read_input_token_cost_above_200k_tokens": 6.6e-7, "litellm_provider": "bedrock_converse", "max_input_tokens": 200000, @@ -1160,6 +1204,7 @@ }, "au.anthropic.claude-sonnet-4-6": { "cache_creation_input_token_cost": 0.000004125, + "cache_creation_input_token_cost_above_1hr": 0.0000066, "cache_read_input_token_cost": 3.3e-7, "input_cost_per_token": 0.0000033, "litellm_provider": "bedrock_converse", @@ -4531,6 +4576,36 @@ "supports_tool_choice": true, "supports_reasoning": true }, + "azure_ai/claude-fable-5": { + "input_cost_per_token": 0.00001, + "output_cost_per_token": 0.00005, + "litellm_provider": "azure_ai", + "max_input_tokens": 1000000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "mode": "chat", + "search_context_cost_per_query": { + "search_context_size_high": 0.01, + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01 + }, + "cache_creation_input_token_cost": 0.0000125, + "cache_creation_input_token_cost_above_1hr": 0.00002, + "cache_read_input_token_cost": 0.000001, + "supports_adaptive_thinking": true, + "supports_assistant_prefill": false, + "supports_computer_use": true, + "supports_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_sampling_params": false, + "supports_tool_choice": true, + "supports_vision": true, + "supports_xhigh_reasoning_effort": true, + "supports_max_reasoning_effort": true + }, "azure_ai/claude-haiku-4-5": { "cache_creation_input_token_cost": 0.00000125, "cache_creation_input_token_cost_above_1hr": 0.000002, @@ -4646,6 +4721,7 @@ "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, + "supports_sampling_params": false, "supports_tool_choice": true, "supports_vision": true, "supports_xhigh_reasoning_effort": true, @@ -4667,6 +4743,7 @@ "cache_creation_input_token_cost": 0.00000625, "cache_creation_input_token_cost_above_1hr": 0.00001, "cache_read_input_token_cost": 5e-7, + "supports_adaptive_thinking": true, "supports_assistant_prefill": false, "supports_computer_use": true, "supports_function_calling": true, @@ -4674,6 +4751,7 @@ "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, + "supports_sampling_params": false, "supports_tool_choice": true, "supports_vision": true, "supports_xhigh_reasoning_effort": true, @@ -4758,6 +4836,19 @@ "supports_function_calling": true, "supports_tool_choice": true }, + "azure_ai/deepseek-v3.1": { + "input_cost_per_token": 0.00000123, + "litellm_provider": "azure_ai", + "max_input_tokens": 131072, + "max_output_tokens": 131072, + "max_tokens": 131072, + "mode": "chat", + "output_cost_per_token": 0.00000494, + "source": "https://azure.microsoft.com/en-us/pricing/details/ai-foundry-models/deepseek/", + "supports_function_calling": true, + "supports_reasoning": true, + "supports_tool_choice": true + }, "azure_ai/deepseek-v3.2": { "input_cost_per_token": 5.8e-7, "litellm_provider": "azure_ai", @@ -4788,6 +4879,32 @@ "supports_reasoning": true, "supports_tool_choice": true }, + "azure_ai/deepseek-v4-flash": { + "input_cost_per_token": 1.9e-7, + "litellm_provider": "azure_ai", + "max_input_tokens": 1000000, + "max_output_tokens": 384000, + "max_tokens": 384000, + "mode": "chat", + "output_cost_per_token": 5.1e-7, + "source": "https://azure.microsoft.com/en-us/pricing/details/ai-foundry-models/deepseek/", + "supports_function_calling": true, + "supports_reasoning": true, + "supports_tool_choice": true + }, + "azure_ai/deepseek-v4-pro": { + "input_cost_per_token": 0.00000174, + "litellm_provider": "azure_ai", + "max_input_tokens": 1000000, + "max_output_tokens": 384000, + "max_tokens": 384000, + "mode": "chat", + "output_cost_per_token": 0.00000348, + "source": "https://azure.microsoft.com/en-us/pricing/details/ai-foundry-models/deepseek/", + "supports_function_calling": true, + "supports_reasoning": true, + "supports_tool_choice": true + }, "azure_ai/global/grok-3": { "input_cost_per_token": 0.000003, "litellm_provider": "azure_ai", @@ -5099,6 +5216,100 @@ "supports_xhigh_reasoning_effort": true, "supports_minimal_reasoning_effort": false }, + "azure_ai/gpt-5.5": { + "cache_read_input_token_cost": 5e-7, + "cache_read_input_token_cost_above_272k_tokens": 0.000001, + "cache_read_input_token_cost_priority": 0.000001, + "cache_read_input_token_cost_above_272k_tokens_priority": 0.000002, + "input_cost_per_token": 0.000005, + "input_cost_per_token_above_272k_tokens": 0.00001, + "input_cost_per_token_priority": 0.00001, + "input_cost_per_token_above_272k_tokens_priority": 0.00002, + "litellm_provider": "azure_ai", + "max_input_tokens": 1050000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "mode": "chat", + "output_cost_per_token": 0.00003, + "output_cost_per_token_above_272k_tokens": 0.000045, + "output_cost_per_token_priority": 0.00006, + "output_cost_per_token_above_272k_tokens_priority": 0.00009, + "source": "https://ai.azure.com/catalog/models/gpt-5.5", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/batch", + "/v1/responses" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], + "supports_function_calling": true, + "supports_native_streaming": true, + "supports_parallel_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_service_tier": true, + "supports_vision": true, + "supports_web_search": true, + "supports_none_reasoning_effort": true, + "supports_xhigh_reasoning_effort": true, + "supports_minimal_reasoning_effort": false + }, + "azure_ai/gpt-5.5-2026-04-23": { + "cache_read_input_token_cost": 5e-7, + "cache_read_input_token_cost_above_272k_tokens": 0.000001, + "cache_read_input_token_cost_priority": 0.000001, + "cache_read_input_token_cost_above_272k_tokens_priority": 0.000002, + "input_cost_per_token": 0.000005, + "input_cost_per_token_above_272k_tokens": 0.00001, + "input_cost_per_token_priority": 0.00001, + "input_cost_per_token_above_272k_tokens_priority": 0.00002, + "litellm_provider": "azure_ai", + "max_input_tokens": 1050000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "mode": "chat", + "output_cost_per_token": 0.00003, + "output_cost_per_token_above_272k_tokens": 0.000045, + "output_cost_per_token_priority": 0.00006, + "output_cost_per_token_above_272k_tokens_priority": 0.00009, + "source": "https://ai.azure.com/catalog/models/gpt-5.5", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/batch", + "/v1/responses" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], + "supports_function_calling": true, + "supports_native_streaming": true, + "supports_parallel_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_service_tier": true, + "supports_vision": true, + "supports_web_search": true, + "supports_none_reasoning_effort": true, + "supports_xhigh_reasoning_effort": true, + "supports_minimal_reasoning_effort": false + }, "azure_ai/gpt-oss-120b": { "input_cost_per_token": 1.5e-7, "output_cost_per_token": 6e-7, @@ -5260,6 +5471,27 @@ "supports_video_input": true, "supports_vision": true }, + "azure_ai/kimi-k2.6": { + "input_cost_per_token": 9.5e-7, + "litellm_provider": "azure_ai", + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "max_tokens": 262144, + "mode": "chat", + "output_cost_per_token": 0.000004, + "source": "https://techcommunity.microsoft.com/blog/azure-ai-foundry-blog/introducing-kimi-k2-6-in-microsoft-foundry/4513125", + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], + "supports_function_calling": true, + "supports_reasoning": true, + "supports_tool_choice": true, + "supports_vision": true + }, "azure_ai/ministral-3b": { "input_cost_per_token": 4e-8, "litellm_provider": "azure_ai", @@ -6729,6 +6961,7 @@ }, "bedrock/us-gov-east-1/anthropic.claude-haiku-4-5-20251001-v1:0": { "cache_creation_input_token_cost": 0.0000015, + "cache_creation_input_token_cost_above_1hr": 0.0000024, "cache_read_input_token_cost": 1.2e-7, "input_cost_per_token": 0.0000012, "litellm_provider": "bedrock", @@ -6750,15 +6983,16 @@ "supports_pdf_input": true }, "bedrock/us-gov-east-1/anthropic.claude-sonnet-4-5-20250929-v1:0": { - "cache_creation_input_token_cost": 0.000004125, - "cache_read_input_token_cost": 3.3e-7, - "input_cost_per_token": 0.0000033, + "cache_creation_input_token_cost": 0.0000045, + "cache_creation_input_token_cost_above_1hr": 0.0000072, + "cache_read_input_token_cost": 3.6e-7, + "input_cost_per_token": 0.0000036, "litellm_provider": "bedrock", "max_input_tokens": 200000, "max_output_tokens": 8192, "max_tokens": 8192, "mode": "chat", - "output_cost_per_token": 0.0000165, + "output_cost_per_token": 0.000018, "supports_assistant_prefill": true, "supports_computer_use": true, "supports_function_calling": true, @@ -6771,15 +7005,16 @@ "supports_native_structured_output": true }, "bedrock/us-gov-east-1/claude-sonnet-4-5-20250929-v1:0": { - "cache_creation_input_token_cost": 0.000004125, - "cache_read_input_token_cost": 3.3e-7, - "input_cost_per_token": 0.0000033, + "cache_creation_input_token_cost": 0.0000045, + "cache_creation_input_token_cost_above_1hr": 0.0000072, + "cache_read_input_token_cost": 3.6e-7, + "input_cost_per_token": 0.0000036, "litellm_provider": "bedrock", "max_input_tokens": 200000, "max_output_tokens": 8192, "max_tokens": 8192, "mode": "chat", - "output_cost_per_token": 0.0000165, + "output_cost_per_token": 0.000018, "supports_assistant_prefill": true, "supports_computer_use": true, "supports_function_calling": true, @@ -6906,6 +7141,7 @@ }, "bedrock/us-gov-west-1/anthropic.claude-haiku-4-5-20251001-v1:0": { "cache_creation_input_token_cost": 0.0000015, + "cache_creation_input_token_cost_above_1hr": 0.0000024, "cache_read_input_token_cost": 1.2e-7, "input_cost_per_token": 0.0000012, "litellm_provider": "bedrock", @@ -6927,15 +7163,16 @@ "supports_pdf_input": true }, "bedrock/us-gov-west-1/anthropic.claude-sonnet-4-5-20250929-v1:0": { - "cache_creation_input_token_cost": 0.000004125, - "cache_read_input_token_cost": 3.3e-7, - "input_cost_per_token": 0.0000033, + "cache_creation_input_token_cost": 0.0000045, + "cache_creation_input_token_cost_above_1hr": 0.0000072, + "cache_read_input_token_cost": 3.6e-7, + "input_cost_per_token": 0.0000036, "litellm_provider": "bedrock", "max_input_tokens": 200000, "max_output_tokens": 8192, "max_tokens": 8192, "mode": "chat", - "output_cost_per_token": 0.0000165, + "output_cost_per_token": 0.000018, "supports_assistant_prefill": true, "supports_computer_use": true, "supports_function_calling": true, @@ -6948,15 +7185,16 @@ "supports_native_structured_output": true }, "bedrock/us-gov-west-1/claude-sonnet-4-5-20250929-v1:0": { - "cache_creation_input_token_cost": 0.000004125, - "cache_read_input_token_cost": 3.3e-7, - "input_cost_per_token": 0.0000033, + "cache_creation_input_token_cost": 0.0000045, + "cache_creation_input_token_cost_above_1hr": 0.0000072, + "cache_read_input_token_cost": 3.6e-7, + "input_cost_per_token": 0.0000036, "litellm_provider": "bedrock", "max_input_tokens": 200000, "max_output_tokens": 8192, "max_tokens": 8192, "mode": "chat", - "output_cost_per_token": 0.0000165, + "output_cost_per_token": 0.000018, "supports_assistant_prefill": true, "supports_computer_use": true, "supports_function_calling": true, @@ -7232,6 +7470,63 @@ "supports_response_schema": true, "supports_tool_choice": true }, + "bedrock_mantle/google.gemma-4-26b-a4b": { + "input_cost_per_token": 1.3e-7, + "output_cost_per_token": 4e-7, + "litellm_provider": "bedrock_mantle", + "max_input_tokens": 256000, + "max_output_tokens": 256000, + "max_tokens": 256000, + "mode": "chat", + "use_openai_responses_path": true, + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/responses" + ], + "supports_function_calling": true, + "supports_parallel_function_calling": false, + "supports_reasoning": true, + "supports_tool_choice": true, + "supports_vision": true + }, + "bedrock_mantle/google.gemma-4-31b": { + "input_cost_per_token": 1.4e-7, + "output_cost_per_token": 4e-7, + "litellm_provider": "bedrock_mantle", + "max_input_tokens": 256000, + "max_output_tokens": 256000, + "max_tokens": 256000, + "mode": "chat", + "use_openai_responses_path": true, + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/responses" + ], + "supports_function_calling": true, + "supports_parallel_function_calling": false, + "supports_reasoning": true, + "supports_tool_choice": true, + "supports_vision": true + }, + "bedrock_mantle/google.gemma-4-e2b": { + "input_cost_per_token": 4e-8, + "output_cost_per_token": 8e-8, + "litellm_provider": "bedrock_mantle", + "max_input_tokens": 128000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "mode": "chat", + "use_openai_responses_path": true, + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/responses" + ], + "supports_function_calling": true, + "supports_parallel_function_calling": false, + "supports_reasoning": true, + "supports_tool_choice": true, + "supports_vision": true + }, "bedrock_mantle/openai.gpt-oss-120b": { "input_cost_per_token": 1.5e-7, "output_cost_per_token": 6e-7, @@ -7240,6 +7535,10 @@ "max_output_tokens": 32768, "max_tokens": 32768, "mode": "chat", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/responses" + ], "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_reasoning": true, @@ -7254,6 +7553,10 @@ "max_output_tokens": 32768, "max_tokens": 32768, "mode": "chat", + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/responses" + ], "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_reasoning": true, @@ -7268,6 +7571,9 @@ "max_output_tokens": 65536, "max_tokens": 65536, "mode": "chat", + "supported_endpoints": [ + "/v1/chat/completions" + ], "supports_function_calling": true, "supports_reasoning": true, "supports_response_schema": true, @@ -7281,6 +7587,9 @@ "max_output_tokens": 65536, "max_tokens": 65536, "mode": "chat", + "supported_endpoints": [ + "/v1/chat/completions" + ], "supports_function_calling": true, "supports_reasoning": true, "supports_response_schema": true, @@ -7579,6 +7888,40 @@ "supports_vision": true, "supports_web_search": true }, + "claude-fable-5": { + "cache_creation_input_token_cost": 0.0000125, + "cache_creation_input_token_cost_above_1hr": 0.00002, + "cache_read_input_token_cost": 0.000001, + "input_cost_per_token": 0.00001, + "litellm_provider": "anthropic", + "max_input_tokens": 1000000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "mode": "chat", + "output_cost_per_token": 0.00005, + "search_context_cost_per_query": { + "search_context_size_high": 0.01, + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01 + }, + "supports_adaptive_thinking": true, + "supports_assistant_prefill": false, + "supports_computer_use": true, + "supports_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_sampling_params": false, + "supports_tool_choice": true, + "supports_vision": true, + "supports_xhigh_reasoning_effort": true, + "supports_max_reasoning_effort": true, + "provider_specific_entry": { + "us": 1.1 + }, + "supports_output_config": true + }, "claude-haiku-4-5": { "cache_creation_input_token_cost": 0.00000125, "cache_creation_input_token_cost_above_1hr": 0.000002, @@ -7845,6 +8188,7 @@ "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, + "supports_sampling_params": false, "supports_tool_choice": true, "supports_vision": true, "supports_xhigh_reasoning_effort": true, @@ -7879,6 +8223,7 @@ "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, + "supports_sampling_params": false, "supports_tool_choice": true, "supports_vision": true, "supports_xhigh_reasoning_effort": true, @@ -7913,6 +8258,7 @@ "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, + "supports_sampling_params": false, "supports_tool_choice": true, "supports_vision": true, "supports_xhigh_reasoning_effort": true, @@ -7956,6 +8302,8 @@ }, "claude-sonnet-4-5": { "cache_creation_input_token_cost": 0.00000375, + "cache_creation_input_token_cost_above_1hr": 0.000006, + "cache_creation_input_token_cost_above_1hr_above_200k_tokens": 0.000012, "cache_read_input_token_cost": 3e-7, "input_cost_per_token": 0.000003, "input_cost_per_token_above_200k_tokens": 0.000006, @@ -7985,6 +8333,8 @@ }, "claude-sonnet-4-5-20250929": { "cache_creation_input_token_cost": 0.00000375, + "cache_creation_input_token_cost_above_1hr": 0.000006, + "cache_creation_input_token_cost_above_1hr_above_200k_tokens": 0.000012, "cache_read_input_token_cost": 3e-7, "input_cost_per_token": 0.000003, "input_cost_per_token_above_200k_tokens": 0.000006, @@ -8015,6 +8365,8 @@ }, "claude-sonnet-4-5-20250929-v1:0": { "cache_creation_input_token_cost": 0.00000375, + "cache_creation_input_token_cost_above_1hr": 0.000006, + "cache_creation_input_token_cost_above_1hr_above_200k_tokens": 0.000012, "cache_read_input_token_cost": 3e-7, "input_cost_per_token": 0.000003, "input_cost_per_token_above_200k_tokens": 0.000006, @@ -8039,6 +8391,7 @@ }, "claude-sonnet-4-6": { "cache_creation_input_token_cost": 0.00000375, + "cache_creation_input_token_cost_above_1hr": 0.000006, "cache_read_input_token_cost": 3e-7, "input_cost_per_token": 0.000003, "litellm_provider": "anthropic", @@ -8231,13 +8584,13 @@ "supports_tool_choice": true }, "command-r7b-12-2024": { - "input_cost_per_token": 1.5e-7, + "input_cost_per_token": 3.75e-8, "litellm_provider": "cohere_chat", "max_input_tokens": 128000, "max_output_tokens": 4096, "max_tokens": 4096, "mode": "chat", - "output_cost_per_token": 3.75e-8, + "output_cost_per_token": 1.5e-7, "source": "https://docs.cohere.com/v2/docs/command-r7b", "supports_function_calling": true, "supports_tool_choice": true @@ -9980,7 +10333,8 @@ "litellm_provider": "deepinfra", "mode": "chat", "supports_tool_choice": true, - "supports_function_calling": true + "supports_function_calling": true, + "supports_image_size": false }, "deepinfra/google/gemini-2.5-pro": { "max_tokens": 1000000, @@ -10377,6 +10731,56 @@ "supports_reasoning": true, "supports_tool_choice": true }, + "deepseek-v4-flash": { + "cache_creation_input_token_cost": 0, + "cache_read_input_token_cost": 2.8e-9, + "input_cost_per_token": 1.4e-7, + "input_cost_per_token_cache_hit": 2.8e-9, + "litellm_provider": "deepseek", + "max_input_tokens": 1000000, + "max_output_tokens": 8192, + "max_tokens": 8192, + "mode": "chat", + "output_cost_per_token": 2.8e-7, + "source": "https://api-docs.deepseek.com/quick_start/pricing", + "supported_endpoints": [ + "/v1/chat/completions" + ], + "supports_assistant_prefill": true, + "supports_function_calling": true, + "supports_native_streaming": true, + "supports_parallel_function_calling": true, + "supports_prompt_caching": true, + "supports_response_schema": true, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_vision": false + }, + "deepseek-v4-pro": { + "cache_creation_input_token_cost": 0, + "cache_read_input_token_cost": 3.625e-9, + "input_cost_per_token": 4.35e-7, + "input_cost_per_token_cache_hit": 3.625e-9, + "litellm_provider": "deepseek", + "max_input_tokens": 1000000, + "max_output_tokens": 8192, + "max_tokens": 8192, + "mode": "chat", + "output_cost_per_token": 8.7e-7, + "source": "https://api-docs.deepseek.com/quick_start/pricing", + "supported_endpoints": [ + "/v1/chat/completions" + ], + "supports_assistant_prefill": true, + "supports_function_calling": true, + "supports_native_streaming": true, + "supports_parallel_function_calling": true, + "supports_prompt_caching": true, + "supports_response_schema": true, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_vision": false + }, "deepseek.v3-v1:0": { "input_cost_per_token": 5.8e-7, "litellm_provider": "bedrock_converse", @@ -10511,6 +10915,56 @@ "supports_reasoning": true, "supports_tool_choice": true }, + "deepseek/deepseek-v4-flash": { + "cache_creation_input_token_cost": 0, + "cache_read_input_token_cost": 2.8e-9, + "input_cost_per_token": 1.4e-7, + "input_cost_per_token_cache_hit": 2.8e-9, + "litellm_provider": "deepseek", + "max_input_tokens": 1000000, + "max_output_tokens": 8192, + "max_tokens": 8192, + "mode": "chat", + "output_cost_per_token": 2.8e-7, + "source": "https://api-docs.deepseek.com/quick_start/pricing", + "supported_endpoints": [ + "/v1/chat/completions" + ], + "supports_assistant_prefill": true, + "supports_function_calling": true, + "supports_native_streaming": true, + "supports_parallel_function_calling": true, + "supports_prompt_caching": true, + "supports_response_schema": true, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_vision": false + }, + "deepseek/deepseek-v4-pro": { + "cache_creation_input_token_cost": 0, + "cache_read_input_token_cost": 3.625e-9, + "input_cost_per_token": 4.35e-7, + "input_cost_per_token_cache_hit": 3.625e-9, + "litellm_provider": "deepseek", + "max_input_tokens": 1000000, + "max_output_tokens": 8192, + "max_tokens": 8192, + "mode": "chat", + "output_cost_per_token": 8.7e-7, + "source": "https://api-docs.deepseek.com/quick_start/pricing", + "supported_endpoints": [ + "/v1/chat/completions" + ], + "supports_assistant_prefill": true, + "supports_function_calling": true, + "supports_native_streaming": true, + "supports_parallel_function_calling": true, + "supports_prompt_caching": true, + "supports_response_schema": true, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_vision": false + }, "eu.amazon.nova-2-lite-v1:0": { "cache_read_input_token_cost": 8.25e-8, "input_cost_per_token": 3.3e-7, @@ -10707,8 +11161,42 @@ "cache_read_input_token_cost": 3e-7, "cache_creation_input_token_cost": 0.00000375 }, + "eu.anthropic.claude-fable-5": { + "cache_creation_input_token_cost": 0.00001375, + "cache_creation_input_token_cost_above_1hr": 0.000022, + "cache_read_input_token_cost": 0.0000011, + "input_cost_per_token": 0.000011, + "litellm_provider": "bedrock_converse", + "max_input_tokens": 1000000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "mode": "chat", + "output_cost_per_token": 0.000055, + "search_context_cost_per_query": { + "search_context_size_high": 0.01, + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01 + }, + "supports_adaptive_thinking": true, + "supports_assistant_prefill": false, + "supports_computer_use": true, + "supports_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_sampling_params": false, + "supports_tool_choice": true, + "supports_vision": true, + "supports_xhigh_reasoning_effort": true, + "supports_native_structured_output": true, + "supports_max_reasoning_effort": true, + "supports_output_config": true, + "bedrock_output_config_effort_ceiling": "xhigh" + }, "eu.anthropic.claude-haiku-4-5-20251001-v1:0": { "cache_creation_input_token_cost": 0.000001375, + "cache_creation_input_token_cost_above_1hr": 0.0000022, "cache_read_input_token_cost": 1.1e-7, "input_cost_per_token": 0.0000011, "deprecation_date": "2026-10-15", @@ -10810,6 +11298,7 @@ }, "eu.anthropic.claude-opus-4-6-v1": { "cache_creation_input_token_cost": 0.000006875, + "cache_creation_input_token_cost_above_1hr": 0.000011, "cache_read_input_token_cost": 5.5e-7, "input_cost_per_token": 0.0000055, "litellm_provider": "bedrock_converse", @@ -10839,6 +11328,7 @@ }, "eu.anthropic.claude-opus-4-7": { "cache_creation_input_token_cost": 0.000006875, + "cache_creation_input_token_cost_above_1hr": 0.000011, "cache_read_input_token_cost": 5.5e-7, "input_cost_per_token": 0.0000055, "litellm_provider": "bedrock_converse", @@ -10859,6 +11349,7 @@ "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, + "supports_sampling_params": false, "supports_tool_choice": true, "supports_vision": true, "supports_xhigh_reasoning_effort": true, @@ -10883,6 +11374,7 @@ "search_context_size_low": 0.01, "search_context_size_medium": 0.01 }, + "supports_adaptive_thinking": true, "supports_assistant_prefill": false, "supports_computer_use": true, "supports_function_calling": true, @@ -10890,6 +11382,7 @@ "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, + "supports_sampling_params": false, "supports_tool_choice": true, "supports_vision": true, "supports_xhigh_reasoning_effort": true, @@ -10929,11 +11422,13 @@ }, "eu.anthropic.claude-sonnet-4-5-20250929-v1:0": { "cache_creation_input_token_cost": 0.000004125, + "cache_creation_input_token_cost_above_1hr": 0.0000066, "cache_read_input_token_cost": 3.3e-7, "input_cost_per_token": 0.0000033, "input_cost_per_token_above_200k_tokens": 0.0000066, "output_cost_per_token_above_200k_tokens": 0.00002475, "cache_creation_input_token_cost_above_200k_tokens": 0.00000825, + "cache_creation_input_token_cost_above_1hr_above_200k_tokens": 0.0000132, "cache_read_input_token_cost_above_200k_tokens": 6.6e-7, "litellm_provider": "bedrock_converse", "max_input_tokens": 200000, @@ -10959,6 +11454,7 @@ }, "eu.anthropic.claude-sonnet-4-6": { "cache_creation_input_token_cost": 0.000004125, + "cache_creation_input_token_cost_above_1hr": 0.0000066, "cache_read_input_token_cost": 3.3e-7, "input_cost_per_token": 0.0000033, "litellm_provider": "bedrock_converse", @@ -11524,6 +12020,38 @@ "supports_response_schema": true, "supports_tool_choice": true }, + "fireworks_ai/accounts/fireworks/models/deepseek-v4-flash": { + "cache_read_input_token_cost": 2.8e-8, + "input_cost_per_token": 1.4e-7, + "litellm_provider": "fireworks_ai", + "max_input_tokens": 1048576, + "max_output_tokens": 384000, + "max_tokens": 384000, + "mode": "chat", + "output_cost_per_token": 2.8e-7, + "source": "https://docs.fireworks.ai/serverless/pricing", + "supports_function_calling": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_vision": false + }, + "fireworks_ai/accounts/fireworks/models/deepseek-v4-pro": { + "cache_read_input_token_cost": 1.45e-7, + "input_cost_per_token": 0.00000174, + "litellm_provider": "fireworks_ai", + "max_input_tokens": 1048576, + "max_output_tokens": 384000, + "max_tokens": 384000, + "mode": "chat", + "output_cost_per_token": 0.00000348, + "source": "https://docs.fireworks.ai/serverless/pricing", + "supports_function_calling": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_vision": false + }, "fireworks_ai/accounts/fireworks/models/devstral-small-2505": { "max_tokens": 131072, "max_input_tokens": 131072, @@ -11780,43 +12308,64 @@ "input_cost_per_token": 0.0000014, "litellm_provider": "fireworks_ai", "max_input_tokens": 202800, - "max_output_tokens": 202800, - "max_tokens": 202800, + "max_output_tokens": 131072, + "max_tokens": 131072, "mode": "chat", "output_cost_per_token": 0.0000044, - "source": "https://fireworks.ai/models/fireworks/glm-5p1", - "supports_function_calling": false, + "source": "https://docs.fireworks.ai/serverless/pricing", + "supports_function_calling": true, "supports_reasoning": true, - "supports_response_schema": false, - "supports_tool_choice": false + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_vision": false + }, + "fireworks_ai/accounts/fireworks/models/glm-5p2": { + "cache_read_input_token_cost": 2.6e-7, + "input_cost_per_token": 0.0000014, + "litellm_provider": "fireworks_ai", + "max_input_tokens": 1048576, + "max_output_tokens": 131072, + "max_tokens": 131072, + "mode": "chat", + "output_cost_per_token": 0.0000044, + "source": "https://docs.fireworks.ai/serverless/pricing", + "supports_function_calling": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_vision": false }, "fireworks_ai/accounts/fireworks/models/gpt-oss-120b": { + "cache_read_input_token_cost": 1.5e-8, "input_cost_per_token": 1.5e-7, "litellm_provider": "fireworks_ai", "max_input_tokens": 131072, - "max_output_tokens": 131072, - "max_tokens": 131072, + "max_output_tokens": 32768, + "max_tokens": 32768, "mode": "chat", "output_cost_per_token": 6e-7, - "source": "https://fireworks.ai/pricing", + "source": "https://docs.fireworks.ai/serverless/pricing", "supports_function_calling": true, "supports_reasoning": true, "supports_response_schema": true, - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_vision": false }, "fireworks_ai/accounts/fireworks/models/gpt-oss-20b": { - "input_cost_per_token": 5e-8, + "cache_read_input_token_cost": 3.5e-8, + "input_cost_per_token": 7e-8, "litellm_provider": "fireworks_ai", "max_input_tokens": 131072, - "max_output_tokens": 131072, - "max_tokens": 131072, + "max_output_tokens": 32768, + "max_tokens": 32768, "mode": "chat", - "output_cost_per_token": 2e-7, - "source": "https://fireworks.ai/pricing", + "output_cost_per_token": 3e-7, + "source": "https://docs.fireworks.ai/serverless/pricing", "supports_function_calling": true, "supports_reasoning": true, "supports_response_schema": true, - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_vision": false }, "fireworks_ai/accounts/fireworks/models/gpt-oss-safeguard-120b": { "max_tokens": 131072, @@ -11953,6 +12502,38 @@ "supports_response_schema": true, "supports_tool_choice": true }, + "fireworks_ai/accounts/fireworks/models/kimi-k2p6": { + "cache_read_input_token_cost": 1.6e-7, + "input_cost_per_token": 9.5e-7, + "litellm_provider": "fireworks_ai", + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "max_tokens": 262144, + "mode": "chat", + "output_cost_per_token": 0.000004, + "source": "https://docs.fireworks.ai/serverless/pricing", + "supports_function_calling": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_vision": true + }, + "fireworks_ai/accounts/fireworks/models/kimi-k2p7-code": { + "cache_read_input_token_cost": 1.9e-7, + "input_cost_per_token": 9.5e-7, + "litellm_provider": "fireworks_ai", + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "max_tokens": 262144, + "mode": "chat", + "output_cost_per_token": 0.000004, + "source": "https://docs.fireworks.ai/serverless/pricing", + "supports_function_calling": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_vision": true + }, "fireworks_ai/accounts/fireworks/models/llama-guard-2-8b": { "max_tokens": 8192, "max_input_tokens": 8192, @@ -12286,6 +12867,38 @@ "supports_response_schema": true, "supports_tool_choice": true }, + "fireworks_ai/accounts/fireworks/models/minimax-m2p7": { + "cache_read_input_token_cost": 6e-8, + "input_cost_per_token": 3e-7, + "litellm_provider": "fireworks_ai", + "max_input_tokens": 196608, + "max_output_tokens": 196608, + "max_tokens": 196608, + "mode": "chat", + "output_cost_per_token": 0.0000012, + "source": "https://docs.fireworks.ai/serverless/pricing", + "supports_function_calling": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_vision": false + }, + "fireworks_ai/accounts/fireworks/models/minimax-m3": { + "cache_read_input_token_cost": 6e-8, + "input_cost_per_token": 3e-7, + "litellm_provider": "fireworks_ai", + "max_input_tokens": 512000, + "max_output_tokens": 512000, + "max_tokens": 512000, + "mode": "chat", + "output_cost_per_token": 0.0000012, + "source": "https://docs.fireworks.ai/serverless/pricing", + "supports_function_calling": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_vision": true + }, "fireworks_ai/accounts/fireworks/models/ministral-3-14b-instruct-2512": { "max_tokens": 256000, "max_input_tokens": 256000, @@ -13228,6 +13841,22 @@ "litellm_provider": "fireworks_ai", "mode": "chat" }, + "fireworks_ai/accounts/fireworks/models/qwen3p7-plus": { + "cache_read_input_token_cost": 8e-8, + "input_cost_per_token": 4e-7, + "litellm_provider": "fireworks_ai", + "max_input_tokens": 262144, + "max_output_tokens": 65536, + "max_tokens": 65536, + "mode": "chat", + "output_cost_per_token": 0.0000016, + "source": "https://docs.fireworks.ai/serverless/pricing", + "supports_function_calling": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_vision": true + }, "fireworks_ai/accounts/fireworks/models/qwq-32b": { "max_tokens": 131072, "max_input_tokens": 131072, @@ -13376,6 +14005,86 @@ "litellm_provider": "fireworks_ai", "mode": "chat" }, + "fireworks_ai/accounts/fireworks/routers/glm-5p1-fast": { + "cache_read_input_token_cost": 5.2e-7, + "input_cost_per_token": 0.0000028, + "litellm_provider": "fireworks_ai", + "max_input_tokens": 202800, + "max_output_tokens": 131072, + "max_tokens": 131072, + "mode": "chat", + "output_cost_per_token": 0.0000088, + "source": "https://docs.fireworks.ai/serverless/pricing", + "supports_function_calling": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_vision": false + }, + "fireworks_ai/accounts/fireworks/routers/kimi-k2p6-fast": { + "cache_read_input_token_cost": 3e-7, + "input_cost_per_token": 0.000002, + "litellm_provider": "fireworks_ai", + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "max_tokens": 262144, + "mode": "chat", + "output_cost_per_token": 0.000008, + "source": "https://docs.fireworks.ai/serverless/pricing", + "supports_function_calling": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_vision": true + }, + "fireworks_ai/accounts/fireworks/routers/kimi-k2p7-code-fast": { + "cache_read_input_token_cost": 3.8e-7, + "input_cost_per_token": 0.0000019, + "litellm_provider": "fireworks_ai", + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "max_tokens": 262144, + "mode": "chat", + "output_cost_per_token": 0.000008, + "source": "https://docs.fireworks.ai/serverless/pricing", + "supports_function_calling": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_vision": true + }, + "fireworks_ai/deepseek-v4-flash": { + "cache_read_input_token_cost": 2.8e-8, + "input_cost_per_token": 1.4e-7, + "litellm_provider": "fireworks_ai", + "max_input_tokens": 1048576, + "max_output_tokens": 384000, + "max_tokens": 384000, + "mode": "chat", + "output_cost_per_token": 2.8e-7, + "source": "https://docs.fireworks.ai/serverless/pricing", + "supports_function_calling": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_vision": false + }, + "fireworks_ai/deepseek-v4-pro": { + "cache_read_input_token_cost": 1.45e-7, + "input_cost_per_token": 0.00000174, + "litellm_provider": "fireworks_ai", + "max_input_tokens": 1048576, + "max_output_tokens": 384000, + "max_tokens": 384000, + "mode": "chat", + "output_cost_per_token": 0.00000348, + "source": "https://docs.fireworks.ai/serverless/pricing", + "supports_function_calling": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_vision": false + }, "fireworks_ai/glm-4p7": { "cache_read_input_token_cost": 3e-7, "input_cost_per_token": 6e-7, @@ -13396,15 +14105,80 @@ "input_cost_per_token": 0.0000014, "litellm_provider": "fireworks_ai", "max_input_tokens": 202800, - "max_output_tokens": 202800, - "max_tokens": 202800, + "max_output_tokens": 131072, + "max_tokens": 131072, "mode": "chat", "output_cost_per_token": 0.0000044, - "source": "https://fireworks.ai/models/fireworks/glm-5p1", - "supports_function_calling": false, + "source": "https://docs.fireworks.ai/serverless/pricing", + "supports_function_calling": true, "supports_reasoning": true, - "supports_response_schema": false, - "supports_tool_choice": false + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_vision": false + }, + "fireworks_ai/glm-5p1-fast": { + "cache_read_input_token_cost": 5.2e-7, + "input_cost_per_token": 0.0000028, + "litellm_provider": "fireworks_ai", + "max_input_tokens": 202800, + "max_output_tokens": 131072, + "max_tokens": 131072, + "mode": "chat", + "output_cost_per_token": 0.0000088, + "source": "https://docs.fireworks.ai/serverless/pricing", + "supports_function_calling": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_vision": false + }, + "fireworks_ai/glm-5p2": { + "cache_read_input_token_cost": 2.6e-7, + "input_cost_per_token": 0.0000014, + "litellm_provider": "fireworks_ai", + "max_input_tokens": 1048576, + "max_output_tokens": 131072, + "max_tokens": 131072, + "mode": "chat", + "output_cost_per_token": 0.0000044, + "source": "https://docs.fireworks.ai/serverless/pricing", + "supports_function_calling": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_vision": false + }, + "fireworks_ai/gpt-oss-120b": { + "cache_read_input_token_cost": 1.5e-8, + "input_cost_per_token": 1.5e-7, + "litellm_provider": "fireworks_ai", + "max_input_tokens": 131072, + "max_output_tokens": 32768, + "max_tokens": 32768, + "mode": "chat", + "output_cost_per_token": 6e-7, + "source": "https://docs.fireworks.ai/serverless/pricing", + "supports_function_calling": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_vision": false + }, + "fireworks_ai/gpt-oss-20b": { + "cache_read_input_token_cost": 3.5e-8, + "input_cost_per_token": 7e-8, + "litellm_provider": "fireworks_ai", + "max_input_tokens": 131072, + "max_output_tokens": 32768, + "max_tokens": 32768, + "mode": "chat", + "output_cost_per_token": 3e-7, + "source": "https://docs.fireworks.ai/serverless/pricing", + "supports_function_calling": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_vision": false }, "fireworks_ai/kimi-k2p5": { "cache_read_input_token_cost": 1e-7, @@ -13420,6 +14194,70 @@ "supports_response_schema": true, "supports_tool_choice": true }, + "fireworks_ai/kimi-k2p6": { + "cache_read_input_token_cost": 1.6e-7, + "input_cost_per_token": 9.5e-7, + "litellm_provider": "fireworks_ai", + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "max_tokens": 262144, + "mode": "chat", + "output_cost_per_token": 0.000004, + "source": "https://docs.fireworks.ai/serverless/pricing", + "supports_function_calling": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_vision": true + }, + "fireworks_ai/kimi-k2p6-fast": { + "cache_read_input_token_cost": 3e-7, + "input_cost_per_token": 0.000002, + "litellm_provider": "fireworks_ai", + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "max_tokens": 262144, + "mode": "chat", + "output_cost_per_token": 0.000008, + "source": "https://docs.fireworks.ai/serverless/pricing", + "supports_function_calling": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_vision": true + }, + "fireworks_ai/kimi-k2p7-code": { + "cache_read_input_token_cost": 1.9e-7, + "input_cost_per_token": 9.5e-7, + "litellm_provider": "fireworks_ai", + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "max_tokens": 262144, + "mode": "chat", + "output_cost_per_token": 0.000004, + "source": "https://docs.fireworks.ai/serverless/pricing", + "supports_function_calling": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_vision": true + }, + "fireworks_ai/kimi-k2p7-code-fast": { + "cache_read_input_token_cost": 3.8e-7, + "input_cost_per_token": 0.0000019, + "litellm_provider": "fireworks_ai", + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "max_tokens": 262144, + "mode": "chat", + "output_cost_per_token": 0.000008, + "source": "https://docs.fireworks.ai/serverless/pricing", + "supports_function_calling": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_vision": true + }, "fireworks_ai/minimax-m2p1": { "cache_read_input_token_cost": 3e-8, "input_cost_per_token": 3e-7, @@ -13434,6 +14272,54 @@ "supports_response_schema": true, "supports_tool_choice": true }, + "fireworks_ai/minimax-m2p7": { + "cache_read_input_token_cost": 6e-8, + "input_cost_per_token": 3e-7, + "litellm_provider": "fireworks_ai", + "max_input_tokens": 196608, + "max_output_tokens": 196608, + "max_tokens": 196608, + "mode": "chat", + "output_cost_per_token": 0.0000012, + "source": "https://docs.fireworks.ai/serverless/pricing", + "supports_function_calling": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_vision": false + }, + "fireworks_ai/minimax-m3": { + "cache_read_input_token_cost": 6e-8, + "input_cost_per_token": 3e-7, + "litellm_provider": "fireworks_ai", + "max_input_tokens": 512000, + "max_output_tokens": 512000, + "max_tokens": 512000, + "mode": "chat", + "output_cost_per_token": 0.0000012, + "source": "https://docs.fireworks.ai/serverless/pricing", + "supports_function_calling": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_vision": true + }, + "fireworks_ai/qwen3p7-plus": { + "cache_read_input_token_cost": 8e-8, + "input_cost_per_token": 4e-7, + "litellm_provider": "fireworks_ai", + "max_input_tokens": 262144, + "max_output_tokens": 65536, + "max_tokens": 65536, + "mode": "chat", + "output_cost_per_token": 0.0000016, + "source": "https://docs.fireworks.ai/serverless/pricing", + "supports_function_calling": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_vision": true + }, "friendliai/meta-llama-3.1-70b-instruct": { "input_cost_per_token": 6e-7, "litellm_provider": "friendliai", @@ -13893,7 +14779,8 @@ "search_context_size_medium": 0.035, "search_context_size_high": 0.035 }, - "supports_service_tier": true + "supports_service_tier": true, + "supports_image_size": false }, "gemini-2.5-flash-image": { "cache_read_input_token_cost": 3e-8, @@ -13943,7 +14830,8 @@ "supports_vision": true, "supports_web_search": false, "tpm": 8000000, - "supports_service_tier": true + "supports_service_tier": true, + "supports_image_size": false }, "gemini-2.5-flash-lite": { "cache_read_input_token_cost": 1e-8, @@ -13994,7 +14882,8 @@ "search_context_size_medium": 0.035, "search_context_size_high": 0.035 }, - "supports_service_tier": true + "supports_service_tier": true, + "supports_image_size": false }, "gemini-2.5-flash-lite-preview-06-17": { "deprecation_date": "2025-11-18", @@ -14045,7 +14934,8 @@ "search_context_size_low": 0.035, "search_context_size_medium": 0.035, "search_context_size_high": 0.035 - } + }, + "supports_image_size": false }, "gemini-2.5-flash-lite-preview-09-2025": { "cache_read_input_token_cost": 1e-8, @@ -14095,7 +14985,8 @@ "search_context_size_low": 0.035, "search_context_size_medium": 0.035, "search_context_size_high": 0.035 - } + }, + "supports_image_size": false }, "gemini-2.5-flash-native-audio-latest": { "input_cost_per_audio_token": 0.000001, @@ -14217,7 +15108,8 @@ "search_context_size_low": 0.035, "search_context_size_medium": 0.035, "search_context_size_high": 0.035 - } + }, + "supports_image_size": false }, "gemini-2.5-pro": { "cache_read_input_token_cost": 1.25e-7, @@ -15387,7 +16279,8 @@ "search_context_size_medium": 0.035, "search_context_size_high": 0.035 }, - "supports_service_tier": true + "supports_service_tier": true, + "supports_image_size": false }, "gemini/gemini-2.5-flash-image": { "cache_read_input_token_cost": 3e-8, @@ -15443,7 +16336,8 @@ "search_context_size_medium": 0.035, "search_context_size_high": 0.035 }, - "supports_service_tier": true + "supports_service_tier": true, + "supports_image_size": false }, "gemini/gemini-2.5-flash-lite": { "cache_read_input_token_cost": 1e-8, @@ -15496,7 +16390,8 @@ "search_context_size_medium": 0.035, "search_context_size_high": 0.035 }, - "supports_service_tier": true + "supports_service_tier": true, + "supports_image_size": false }, "gemini/gemini-2.5-flash-lite-preview-06-17": { "deprecation_date": "2025-11-18", @@ -15549,7 +16444,8 @@ "search_context_size_low": 0.035, "search_context_size_medium": 0.035, "search_context_size_high": 0.035 - } + }, + "supports_image_size": false }, "gemini/gemini-2.5-flash-lite-preview-09-2025": { "cache_read_input_token_cost": 1e-8, @@ -15601,7 +16497,8 @@ "search_context_size_low": 0.035, "search_context_size_medium": 0.035, "search_context_size_high": 0.035 - } + }, + "supports_image_size": false }, "gemini/gemini-2.5-flash-native-audio-latest": { "input_cost_per_audio_token": 0.000001, @@ -15731,7 +16628,8 @@ "search_context_size_low": 0.035, "search_context_size_medium": 0.035, "search_context_size_high": 0.035 - } + }, + "supports_image_size": false }, "gemini/gemini-2.5-pro": { "cache_read_input_token_cost": 1.25e-7, @@ -17088,6 +17986,38 @@ "supports_response_schema": true, "supports_vision": true }, + "github_copilot/mai-code-1-flash": { + "cache_read_input_token_cost": 7.5e-8, + "input_cost_per_token": 7.5e-7, + "litellm_provider": "github_copilot", + "max_input_tokens": 128000, + "max_output_tokens": 64000, + "max_tokens": 64000, + "mode": "chat", + "output_cost_per_token": 0.0000045, + "supported_endpoints": [ + "/v1/chat/completions" + ], + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_response_schema": true + }, + "github_copilot/mai-code-1-flash-internal": { + "cache_read_input_token_cost": 7.5e-8, + "input_cost_per_token": 7.5e-7, + "litellm_provider": "github_copilot", + "max_input_tokens": 128000, + "max_output_tokens": 64000, + "max_tokens": 64000, + "mode": "chat", + "output_cost_per_token": 0.0000045, + "supported_endpoints": [ + "/v1/chat/completions" + ], + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_response_schema": true + }, "glm-4-7-251222": { "input_cost_per_token": 0, "litellm_provider": "volcengine", @@ -17119,6 +18049,39 @@ "supports_video_input": true, "supports_vision": true }, + "global.anthropic.claude-fable-5": { + "cache_creation_input_token_cost": 0.0000125, + "cache_creation_input_token_cost_above_1hr": 0.00002, + "cache_read_input_token_cost": 0.000001, + "input_cost_per_token": 0.00001, + "litellm_provider": "bedrock_converse", + "max_input_tokens": 1000000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "mode": "chat", + "output_cost_per_token": 0.00005, + "search_context_cost_per_query": { + "search_context_size_high": 0.01, + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01 + }, + "supports_adaptive_thinking": true, + "supports_assistant_prefill": false, + "supports_computer_use": true, + "supports_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_sampling_params": false, + "supports_tool_choice": true, + "supports_vision": true, + "supports_xhigh_reasoning_effort": true, + "supports_native_structured_output": true, + "supports_max_reasoning_effort": true, + "supports_output_config": true, + "bedrock_output_config_effort_ceiling": "xhigh" + }, "global.anthropic.claude-haiku-4-5-20251001-v1:0": { "cache_creation_input_token_cost": 0.00000125, "cache_creation_input_token_cost_above_1hr": 0.000002, @@ -17224,6 +18187,7 @@ "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, + "supports_sampling_params": false, "supports_tool_choice": true, "supports_vision": true, "supports_xhigh_reasoning_effort": true, @@ -17248,6 +18212,7 @@ "search_context_size_low": 0.01, "search_context_size_medium": 0.01 }, + "supports_adaptive_thinking": true, "supports_assistant_prefill": false, "supports_computer_use": true, "supports_function_calling": true, @@ -17255,6 +18220,7 @@ "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, + "supports_sampling_params": false, "supports_tool_choice": true, "supports_vision": true, "supports_xhigh_reasoning_effort": true, @@ -20412,6 +21378,21 @@ "supports_system_messages": true, "supports_tool_choice": true }, + "inception/mercury-2": { + "cache_read_input_token_cost": 2.5e-8, + "input_cost_per_token": 2.5e-7, + "litellm_provider": "inception", + "max_input_tokens": 128000, + "max_output_tokens": 50000, + "max_tokens": 50000, + "mode": "chat", + "output_cost_per_token": 7.5e-7, + "supports_function_calling": true, + "supports_prompt_caching": true, + "supports_response_schema": true, + "supports_system_messages": true, + "supports_tool_choice": true + }, "jamba-1.5": { "input_cost_per_token": 2e-7, "litellm_provider": "ai21", @@ -20504,6 +21485,7 @@ }, "jp.anthropic.claude-haiku-4-5-20251001-v1:0": { "cache_creation_input_token_cost": 0.000001375, + "cache_creation_input_token_cost_above_1hr": 0.0000022, "cache_read_input_token_cost": 1.1e-7, "input_cost_per_token": 0.0000011, "litellm_provider": "bedrock_converse", @@ -20524,13 +21506,46 @@ "supports_vision": true, "supports_native_structured_output": true }, + "jp.anthropic.claude-opus-4-7": { + "cache_creation_input_token_cost": 0.000006875, + "cache_read_input_token_cost": 5.5e-7, + "input_cost_per_token": 0.0000055, + "litellm_provider": "bedrock_converse", + "max_input_tokens": 1000000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "mode": "chat", + "output_cost_per_token": 0.0000275, + "search_context_cost_per_query": { + "search_context_size_high": 0.01, + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01 + }, + "supports_assistant_prefill": false, + "supports_computer_use": true, + "supports_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_sampling_params": false, + "supports_tool_choice": true, + "supports_vision": true, + "supports_xhigh_reasoning_effort": true, + "tool_use_system_prompt_tokens": 346, + "supports_native_structured_output": true, + "supports_max_reasoning_effort": true, + "supports_minimal_reasoning_effort": true + }, "jp.anthropic.claude-sonnet-4-5-20250929-v1:0": { "cache_creation_input_token_cost": 0.000004125, + "cache_creation_input_token_cost_above_1hr": 0.0000066, "cache_read_input_token_cost": 3.3e-7, "input_cost_per_token": 0.0000033, "input_cost_per_token_above_200k_tokens": 0.0000066, "output_cost_per_token_above_200k_tokens": 0.00002475, "cache_creation_input_token_cost_above_200k_tokens": 0.00000825, + "cache_creation_input_token_cost_above_1hr_above_200k_tokens": 0.0000132, "cache_read_input_token_cost_above_200k_tokens": 6.6e-7, "litellm_provider": "bedrock_converse", "max_input_tokens": 200000, @@ -20556,6 +21571,7 @@ }, "jp.anthropic.claude-sonnet-4-6": { "cache_creation_input_token_cost": 0.000004125, + "cache_creation_input_token_cost_above_1hr": 0.0000066, "cache_read_input_token_cost": 3.3e-7, "input_cost_per_token": 0.0000033, "litellm_provider": "bedrock_converse", @@ -20921,6 +21937,165 @@ "supports_response_schema": true, "supports_tool_choice": true }, + "libertai/deepseek-v4-flash": { + "max_tokens": 200000, + "max_input_tokens": 200000, + "max_output_tokens": 200000, + "input_cost_per_token": 2.5e-7, + "output_cost_per_token": 0.00000175, + "litellm_provider": "libertai", + "mode": "chat", + "supports_function_calling": true, + "supports_tool_choice": true, + "supports_system_messages": true, + "supports_vision": false, + "source": "https://docs.libertai.io/apis/text/" + }, + "libertai/deepseek-v4-flash-thinking": { + "max_tokens": 200000, + "max_input_tokens": 200000, + "max_output_tokens": 200000, + "input_cost_per_token": 2.5e-7, + "output_cost_per_token": 0.00000175, + "litellm_provider": "libertai", + "mode": "chat", + "supports_function_calling": true, + "supports_tool_choice": true, + "supports_system_messages": true, + "supports_vision": false, + "supports_reasoning": true, + "source": "https://docs.libertai.io/apis/text/" + }, + "libertai/gemma-4-31b-it": { + "max_tokens": 262144, + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "input_cost_per_token": 1.5e-7, + "output_cost_per_token": 4e-7, + "litellm_provider": "libertai", + "mode": "chat", + "supports_function_calling": true, + "supports_tool_choice": true, + "supports_system_messages": true, + "supports_vision": true, + "source": "https://docs.libertai.io/apis/text/" + }, + "libertai/gemma-4-31b-it-thinking": { + "max_tokens": 262144, + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "input_cost_per_token": 1.5e-7, + "output_cost_per_token": 4e-7, + "litellm_provider": "libertai", + "mode": "chat", + "supports_function_calling": true, + "supports_tool_choice": true, + "supports_system_messages": true, + "supports_vision": true, + "supports_reasoning": true, + "source": "https://docs.libertai.io/apis/text/" + }, + "libertai/hermes-3-8b-tee": { + "max_tokens": 16000, + "max_input_tokens": 16000, + "max_output_tokens": 16000, + "input_cost_per_token": 1.5e-7, + "output_cost_per_token": 6e-7, + "litellm_provider": "libertai", + "mode": "chat", + "supports_function_calling": true, + "supports_tool_choice": true, + "supports_system_messages": true, + "supports_vision": false, + "source": "https://docs.libertai.io/apis/text/" + }, + "libertai/qwen3.5-122b-a10b": { + "max_tokens": 262144, + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "input_cost_per_token": 2.5e-7, + "output_cost_per_token": 0.00000175, + "litellm_provider": "libertai", + "mode": "chat", + "supports_function_calling": true, + "supports_tool_choice": true, + "supports_system_messages": true, + "supports_vision": true, + "source": "https://docs.libertai.io/apis/text/" + }, + "libertai/qwen3.5-122b-a10b-thinking": { + "max_tokens": 262144, + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "input_cost_per_token": 2.5e-7, + "output_cost_per_token": 0.00000175, + "litellm_provider": "libertai", + "mode": "chat", + "supports_function_calling": true, + "supports_tool_choice": true, + "supports_system_messages": true, + "supports_vision": true, + "supports_reasoning": true, + "source": "https://docs.libertai.io/apis/text/" + }, + "libertai/qwen3.6-27b": { + "max_tokens": 262144, + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "input_cost_per_token": 1.5e-7, + "output_cost_per_token": 5e-7, + "litellm_provider": "libertai", + "mode": "chat", + "supports_function_calling": true, + "supports_tool_choice": true, + "supports_system_messages": true, + "supports_vision": true, + "source": "https://docs.libertai.io/apis/text/" + }, + "libertai/qwen3.6-27b-thinking": { + "max_tokens": 262144, + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "input_cost_per_token": 1.5e-7, + "output_cost_per_token": 5e-7, + "litellm_provider": "libertai", + "mode": "chat", + "supports_function_calling": true, + "supports_tool_choice": true, + "supports_system_messages": true, + "supports_vision": true, + "supports_reasoning": true, + "source": "https://docs.libertai.io/apis/text/" + }, + "libertai/qwen3.6-35b-a3b": { + "max_tokens": 262144, + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "input_cost_per_token": 1.5e-7, + "output_cost_per_token": 5e-7, + "litellm_provider": "libertai", + "mode": "chat", + "supports_function_calling": true, + "supports_tool_choice": true, + "supports_system_messages": true, + "supports_vision": true, + "source": "https://docs.libertai.io/apis/text/" + }, + "libertai/qwen3.6-35b-a3b-thinking": { + "max_tokens": 262144, + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "input_cost_per_token": 1.5e-7, + "output_cost_per_token": 5e-7, + "litellm_provider": "libertai", + "mode": "chat", + "supports_function_calling": true, + "supports_tool_choice": true, + "supports_system_messages": true, + "supports_vision": true, + "supports_reasoning": true, + "source": "https://docs.libertai.io/apis/text/" + }, "llamagate/codellama-7b": { "max_tokens": 4096, "max_input_tokens": 16384, @@ -21448,6 +22623,24 @@ "max_input_tokens": 1000000, "max_output_tokens": 8192 }, + "minimax/MiniMax-M3": { + "input_cost_per_token": 3e-7, + "input_cost_per_token_above_512k_tokens": 6e-7, + "output_cost_per_token": 0.0000012, + "output_cost_per_token_above_512k_tokens": 0.0000024, + "cache_read_input_token_cost": 6e-8, + "cache_read_input_token_cost_above_512k_tokens": 1.2e-7, + "litellm_provider": "minimax", + "mode": "chat", + "supports_function_calling": true, + "supports_tool_choice": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_system_messages": true, + "supports_vision": true, + "max_input_tokens": 1000000, + "max_output_tokens": 128000 + }, "mistral.devstral-2-123b": { "input_cost_per_token": 4e-7, "litellm_provider": "bedrock_converse", @@ -21923,6 +23116,21 @@ "supports_tool_choice": true, "supports_vision": true }, + "mistral/ministral-8b-latest": { + "input_cost_per_token": 1.5e-7, + "litellm_provider": "mistral", + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "max_tokens": 262144, + "mode": "chat", + "output_cost_per_token": 1.5e-7, + "source": "https://mistral.ai/pricing", + "supports_assistant_prefill": true, + "supports_function_calling": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_vision": true + }, "mistral/mistral-large-2402": { "input_cost_per_token": 0.000004, "litellm_provider": "mistral", @@ -22059,6 +23267,21 @@ "supports_tool_choice": true, "supports_vision": true }, + "mistral/mistral-medium-3-5": { + "input_cost_per_token": 0.0000015, + "litellm_provider": "mistral", + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "max_tokens": 262144, + "mode": "chat", + "output_cost_per_token": 0.0000075, + "source": "https://docs.mistral.ai/models/model-cards/mistral-medium-3-5-26-04", + "supports_assistant_prefill": true, + "supports_function_calling": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_vision": true + }, "mistral/mistral-medium-latest": { "input_cost_per_token": 4e-7, "litellm_provider": "mistral", @@ -22260,6 +23483,7 @@ }, "moonshot/kimi-k2-0711-preview": { "cache_read_input_token_cost": 1.5e-7, + "deprecation_date": "2026-05-25", "input_cost_per_token": 6e-7, "litellm_provider": "moonshot", "max_input_tokens": 131072, @@ -22274,6 +23498,7 @@ }, "moonshot/kimi-k2-0905-preview": { "cache_read_input_token_cost": 1.5e-7, + "deprecation_date": "2026-05-25", "input_cost_per_token": 6e-7, "litellm_provider": "moonshot", "max_input_tokens": 262144, @@ -22288,6 +23513,7 @@ }, "moonshot/kimi-k2-thinking": { "cache_read_input_token_cost": 1.5e-7, + "deprecation_date": "2026-05-25", "input_cost_per_token": 6e-7, "litellm_provider": "moonshot", "max_input_tokens": 262144, @@ -22303,6 +23529,7 @@ }, "moonshot/kimi-k2-thinking-turbo": { "cache_read_input_token_cost": 1.5e-7, + "deprecation_date": "2026-05-25", "input_cost_per_token": 0.00000115, "litellm_provider": "moonshot", "max_input_tokens": 262144, @@ -22318,6 +23545,7 @@ }, "moonshot/kimi-k2-turbo-preview": { "cache_read_input_token_cost": 1.5e-7, + "deprecation_date": "2026-05-25", "input_cost_per_token": 0.00000115, "litellm_provider": "moonshot", "max_input_tokens": 262144, @@ -22342,6 +23570,7 @@ "source": "https://platform.moonshot.ai/docs/guide/kimi-k2-5-quickstart", "supports_function_calling": true, "supports_reasoning": true, + "supports_response_schema": true, "supports_tool_choice": true, "supports_video_input": true, "supports_vision": true @@ -22358,12 +23587,14 @@ "source": "https://platform.kimi.ai/docs/pricing/chat-k26", "supports_function_calling": true, "supports_reasoning": true, + "supports_response_schema": true, "supports_tool_choice": true, "supports_video_input": true, "supports_vision": true }, "moonshot/kimi-latest": { "cache_read_input_token_cost": 1.5e-7, + "deprecation_date": "2026-01-28", "input_cost_per_token": 0.000002, "litellm_provider": "moonshot", "max_input_tokens": 131072, @@ -22378,6 +23609,7 @@ }, "moonshot/kimi-latest-128k": { "cache_read_input_token_cost": 1.5e-7, + "deprecation_date": "2026-01-28", "input_cost_per_token": 0.000002, "litellm_provider": "moonshot", "max_input_tokens": 131072, @@ -22392,6 +23624,7 @@ }, "moonshot/kimi-latest-32k": { "cache_read_input_token_cost": 1.5e-7, + "deprecation_date": "2026-01-28", "input_cost_per_token": 0.000001, "litellm_provider": "moonshot", "max_input_tokens": 32768, @@ -22406,6 +23639,7 @@ }, "moonshot/kimi-latest-8k": { "cache_read_input_token_cost": 1.5e-7, + "deprecation_date": "2026-01-28", "input_cost_per_token": 2e-7, "litellm_provider": "moonshot", "max_input_tokens": 8192, @@ -22420,6 +23654,7 @@ }, "moonshot/kimi-thinking-preview": { "cache_read_input_token_cost": 1.5e-7, + "deprecation_date": "2025-11-11", "input_cost_per_token": 6e-7, "litellm_provider": "moonshot", "max_input_tokens": 131072, @@ -22440,9 +23675,11 @@ "output_cost_per_token": 0.000005, "source": "https://platform.moonshot.ai/docs/pricing", "supports_function_calling": true, + "supports_response_schema": true, "supports_tool_choice": true }, "moonshot/moonshot-v1-128k-0430": { + "deprecation_date": "2024-04-30", "input_cost_per_token": 0.000002, "litellm_provider": "moonshot", "max_input_tokens": 131072, @@ -22464,6 +23701,7 @@ "output_cost_per_token": 0.000005, "source": "https://platform.moonshot.ai/docs/pricing", "supports_function_calling": true, + "supports_response_schema": true, "supports_tool_choice": true, "supports_vision": true }, @@ -22477,9 +23715,11 @@ "output_cost_per_token": 0.000003, "source": "https://platform.moonshot.ai/docs/pricing", "supports_function_calling": true, + "supports_response_schema": true, "supports_tool_choice": true }, "moonshot/moonshot-v1-32k-0430": { + "deprecation_date": "2024-04-30", "input_cost_per_token": 0.000001, "litellm_provider": "moonshot", "max_input_tokens": 32768, @@ -22501,6 +23741,7 @@ "output_cost_per_token": 0.000003, "source": "https://platform.moonshot.ai/docs/pricing", "supports_function_calling": true, + "supports_response_schema": true, "supports_tool_choice": true, "supports_vision": true }, @@ -22514,9 +23755,11 @@ "output_cost_per_token": 0.000002, "source": "https://platform.moonshot.ai/docs/pricing", "supports_function_calling": true, + "supports_response_schema": true, "supports_tool_choice": true }, "moonshot/moonshot-v1-8k-0430": { + "deprecation_date": "2024-04-30", "input_cost_per_token": 2e-7, "litellm_provider": "moonshot", "max_input_tokens": 8192, @@ -22538,6 +23781,7 @@ "output_cost_per_token": 0.000002, "source": "https://platform.moonshot.ai/docs/pricing", "supports_function_calling": true, + "supports_response_schema": true, "supports_tool_choice": true, "supports_vision": true }, @@ -22551,6 +23795,7 @@ "output_cost_per_token": 0.000005, "source": "https://platform.moonshot.ai/docs/pricing", "supports_function_calling": true, + "supports_response_schema": true, "supports_tool_choice": true }, "moonshotai.kimi-k2.5": { @@ -24619,7 +25864,8 @@ "supports_function_calling": true, "supports_response_schema": true, "supports_vision": true, - "supports_native_streaming": true + "supports_native_streaming": true, + "supports_image_size": false }, "oci/google.gemini-2.5-flash-lite": { "input_cost_per_token": 7.5e-8, @@ -24633,7 +25879,8 @@ "supports_function_calling": true, "supports_response_schema": true, "supports_vision": true, - "supports_native_streaming": true + "supports_native_streaming": true, + "supports_image_size": false }, "oci/google.gemini-2.5-pro": { "input_cost_per_token": 0.00000125, @@ -25572,7 +26819,8 @@ "supports_response_schema": true, "supports_system_messages": true, "supports_tool_choice": true, - "supports_vision": true + "supports_vision": true, + "supports_image_size": false }, "openrouter/google/gemini-2.5-pro": { "input_cost_per_audio_token": 7e-7, @@ -27207,6 +28455,84 @@ "mode": "chat", "output_cost_per_token": 2.8e-7 }, + "pinstripes/ps/deepseek-v4-flash": { + "max_tokens": 163840, + "max_input_tokens": 163840, + "max_output_tokens": 163840, + "input_cost_per_token": 1e-7, + "output_cost_per_token": 2e-7, + "litellm_provider": "pinstripes", + "mode": "chat", + "supports_function_calling": true, + "supports_assistant_prefill": true, + "supports_reasoning": true, + "source": "https://pinstripes.io/pricing" + }, + "pinstripes/ps/glm-4.5-air": { + "max_tokens": 128000, + "max_input_tokens": 128000, + "max_output_tokens": 128000, + "input_cost_per_token": 1.25e-7, + "output_cost_per_token": 4.5e-7, + "litellm_provider": "pinstripes", + "mode": "chat", + "supports_function_calling": true, + "supports_assistant_prefill": true, + "supports_reasoning": true, + "source": "https://pinstripes.io/pricing" + }, + "pinstripes/ps/minimax-m2.7": { + "max_tokens": 1000192, + "max_input_tokens": 1000192, + "max_output_tokens": 1000192, + "input_cost_per_token": 2.55e-7, + "output_cost_per_token": 5.5e-7, + "litellm_provider": "pinstripes", + "mode": "chat", + "supports_function_calling": true, + "supports_assistant_prefill": true, + "supports_reasoning": false, + "source": "https://pinstripes.io/pricing" + }, + "pinstripes/ps/qwen3-30b-a3b": { + "max_tokens": 131072, + "max_input_tokens": 131072, + "max_output_tokens": 131072, + "input_cost_per_token": 9e-8, + "output_cost_per_token": 2e-7, + "litellm_provider": "pinstripes", + "mode": "chat", + "supports_function_calling": true, + "supports_assistant_prefill": true, + "supports_reasoning": true, + "source": "https://pinstripes.io/pricing" + }, + "pinstripes/ps/qwen3-coder-30b-a3b": { + "max_tokens": 131072, + "max_input_tokens": 131072, + "max_output_tokens": 131072, + "input_cost_per_token": 3e-7, + "output_cost_per_token": 6e-7, + "litellm_provider": "pinstripes", + "mode": "chat", + "supports_function_calling": true, + "supports_assistant_prefill": true, + "supports_reasoning": false, + "source": "https://pinstripes.io/pricing" + }, + "pinstripes/ps/qwen3.6-35b-a3b": { + "max_tokens": 131072, + "max_input_tokens": 131072, + "max_output_tokens": 131072, + "input_cost_per_token": 1.4e-7, + "output_cost_per_token": 4.5e-7, + "litellm_provider": "pinstripes", + "mode": "chat", + "supports_function_calling": true, + "supports_assistant_prefill": true, + "supports_reasoning": true, + "source": "https://pinstripes.io/pricing" + }, "publicai/BSC-LT/ALIA-40b-instruct_Q8_0": { "input_cost_per_token": 0, "litellm_provider": "publicai", @@ -27530,7 +28856,8 @@ "supports_vision": true, "supports_system_messages": true, "supports_tool_choice": true, - "supports_response_schema": true + "supports_response_schema": true, + "supports_image_size": false }, "replicate/google/gemini-3-pro": { "input_cost_per_token": 0.000002, @@ -28079,21 +29406,278 @@ "output_cost_per_token": 0, "supports_reasoning": true }, - "snowflake/claude-3-5-sonnet": { - "litellm_provider": "snowflake", - "max_input_tokens": 18000, + "scaleway/google/gemma-3-27b-it": { + "input_cost_per_token": 2.5e-7, + "litellm_provider": "scaleway", + "max_input_tokens": 40000, "max_output_tokens": 8192, "max_tokens": 8192, "mode": "chat", - "supports_computer_use": true + "output_cost_per_token": 5e-7, + "supports_function_calling": true, + "supports_vision": true + }, + "scaleway/google/gemma-4-26b-a4b-it": { + "input_cost_per_token": 2.5e-7, + "litellm_provider": "scaleway", + "max_input_tokens": 256000, + "max_output_tokens": 32768, + "max_tokens": 32768, + "mode": "chat", + "output_cost_per_token": 5e-7, + "supports_function_calling": true, + "supports_reasoning": true, + "supports_vision": true + }, + "scaleway/hcompany/holo2-30b-a3b": { + "input_cost_per_token": 3e-7, + "litellm_provider": "scaleway", + "max_input_tokens": 22000, + "max_output_tokens": 16384, + "max_tokens": 16384, + "mode": "chat", + "output_cost_per_token": 7e-7, + "supports_reasoning": true, + "supports_vision": true + }, + "scaleway/meta/llama-3.3-70b-instruct": { + "input_cost_per_token": 9e-7, + "litellm_provider": "scaleway", + "max_input_tokens": 128000, + "max_output_tokens": 16384, + "max_tokens": 16384, + "mode": "chat", + "output_cost_per_token": 9e-7, + "supports_function_calling": true + }, + "scaleway/mistralai/devstral-2-123b-instruct-2512": { + "input_cost_per_token": 4e-7, + "litellm_provider": "scaleway", + "max_input_tokens": 200000, + "max_output_tokens": 16384, + "max_tokens": 16384, + "mode": "chat", + "output_cost_per_token": 0.000002, + "supports_function_calling": true + }, + "scaleway/mistralai/mistral-medium-3.5-128b": { + "input_cost_per_token": 0.0000015, + "litellm_provider": "scaleway", + "max_input_tokens": 256000, + "max_output_tokens": 16384, + "max_tokens": 16384, + "mode": "chat", + "output_cost_per_token": 0.0000075, + "supports_reasoning": true, + "supports_function_calling": true, + "supports_vision": true, + "supports_tool_choice": true + }, + "scaleway/mistralai/mistral-small-3.2-24b-instruct-2506": { + "input_cost_per_token": 1.5e-7, + "litellm_provider": "scaleway", + "max_input_tokens": 128000, + "max_output_tokens": 32768, + "max_tokens": 32768, + "mode": "chat", + "output_cost_per_token": 3.5e-7, + "supports_function_calling": true, + "supports_vision": true + }, + "scaleway/mistralai/pixtral-12b-2409": { + "input_cost_per_token": 2e-7, + "litellm_provider": "scaleway", + "max_input_tokens": 128000, + "max_output_tokens": 4096, + "max_tokens": 4096, + "mode": "chat", + "output_cost_per_token": 2e-7, + "supports_vision": true, + "supports_function_calling": true + }, + "scaleway/mistralai/voxtral-small-24b-2507": { + "input_cost_per_audio_token": 1.5e-7, + "input_cost_per_token": 1.5e-7, + "litellm_provider": "scaleway", + "max_input_tokens": 32000, + "max_output_tokens": 16384, + "max_tokens": 16384, + "mode": "chat", + "output_cost_per_token": 3.5e-7, + "supports_audio_input": true + }, + "scaleway/openai/gpt-oss-120b": { + "input_cost_per_token": 1.5e-7, + "litellm_provider": "scaleway", + "max_input_tokens": 128000, + "max_output_tokens": 32768, + "max_tokens": 32768, + "mode": "chat", + "output_cost_per_token": 6e-7, + "supports_function_calling": true + }, + "scaleway/qwen/qwen3-235b-a22b-instruct-2507": { + "input_cost_per_token": 7.5e-7, + "litellm_provider": "scaleway", + "max_input_tokens": 256000, + "max_output_tokens": 16384, + "max_tokens": 16384, + "mode": "chat", + "output_cost_per_token": 0.00000225, + "supports_function_calling": true + }, + "scaleway/qwen/qwen3-coder-30b-a3b-instruct": { + "input_cost_per_token": 2e-7, + "litellm_provider": "scaleway", + "max_input_tokens": 128000, + "max_output_tokens": 32768, + "max_tokens": 32768, + "mode": "chat", + "output_cost_per_token": 8e-7, + "supports_function_calling": true + }, + "scaleway/qwen/qwen3.5-397b-a17b": { + "input_cost_per_token": 6e-7, + "litellm_provider": "scaleway", + "max_input_tokens": 256000, + "max_output_tokens": 16384, + "max_tokens": 16384, + "mode": "chat", + "output_cost_per_token": 0.0000036, + "supports_function_calling": true, + "supports_reasoning": true, + "supports_vision": true + }, + "scaleway/qwen/qwen3.6-35b-a3b": { + "input_cost_per_token": 2.5e-7, + "litellm_provider": "scaleway", + "max_input_tokens": 256000, + "max_output_tokens": 16384, + "max_tokens": 16384, + "mode": "chat", + "output_cost_per_token": 0.0000015, + "supports_function_calling": true, + "supports_vision": true, + "supports_reasoning": true + }, + "snowflake/claude-3-5-sonnet": { + "litellm_provider": "snowflake", + "max_input_tokens": 200000, + "max_output_tokens": 16384, + "max_tokens": 16384, + "mode": "chat", + "input_cost_per_token": 0.000003, + "output_cost_per_token": 0.000015, + "cache_read_input_token_cost": 3e-7, + "supports_computer_use": true, + "supports_function_calling": true, + "supports_vision": true, + "supports_prompt_caching": true, + "supports_system_messages": true, + "supports_response_schema": true + }, + "snowflake/claude-3-7-sonnet": { + "max_tokens": 16384, + "max_input_tokens": 200000, + "max_output_tokens": 16384, + "input_cost_per_token": 0.000003, + "output_cost_per_token": 0.000015, + "cache_read_input_token_cost": 3e-7, + "litellm_provider": "snowflake", + "mode": "chat", + "supports_function_calling": true, + "supports_vision": true, + "supports_prompt_caching": true, + "supports_system_messages": true, + "supports_reasoning": true, + "supports_response_schema": true + }, + "snowflake/claude-4-opus": { + "max_tokens": 16384, + "max_input_tokens": 200000, + "max_output_tokens": 16384, + "input_cost_per_token": 0.000005, + "output_cost_per_token": 0.000025, + "cache_read_input_token_cost": 5e-7, + "litellm_provider": "snowflake", + "mode": "chat", + "supports_function_calling": true, + "supports_vision": true, + "supports_prompt_caching": true, + "supports_system_messages": true, + "supports_reasoning": true, + "supports_response_schema": true + }, + "snowflake/claude-4-sonnet": { + "max_tokens": 16384, + "max_input_tokens": 200000, + "max_output_tokens": 16384, + "input_cost_per_token": 0.000003, + "output_cost_per_token": 0.000015, + "cache_read_input_token_cost": 3e-7, + "litellm_provider": "snowflake", + "mode": "chat", + "supports_function_calling": true, + "supports_vision": true, + "supports_prompt_caching": true, + "supports_system_messages": true, + "supports_response_schema": true + }, + "snowflake/claude-haiku-4-5": { + "max_tokens": 16384, + "max_input_tokens": 200000, + "max_output_tokens": 16384, + "input_cost_per_token": 0.000001, + "output_cost_per_token": 0.000005, + "cache_read_input_token_cost": 1e-7, + "litellm_provider": "snowflake", + "mode": "chat", + "supports_function_calling": true, + "supports_vision": true, + "supports_prompt_caching": true, + "supports_system_messages": true, + "supports_response_schema": true + }, + "snowflake/claude-sonnet-4-5": { + "max_tokens": 16384, + "max_input_tokens": 200000, + "max_output_tokens": 16384, + "input_cost_per_token": 0.000003, + "output_cost_per_token": 0.000015, + "cache_read_input_token_cost": 3e-7, + "litellm_provider": "snowflake", + "mode": "chat", + "supports_function_calling": true, + "supports_vision": true, + "supports_prompt_caching": true, + "supports_system_messages": true, + "supports_response_schema": true + }, + "snowflake/claude-sonnet-4-6": { + "max_tokens": 16384, + "max_input_tokens": 200000, + "max_output_tokens": 16384, + "input_cost_per_token": 0.000003, + "output_cost_per_token": 0.000015, + "cache_read_input_token_cost": 3e-7, + "litellm_provider": "snowflake", + "mode": "chat", + "supports_function_calling": true, + "supports_vision": true, + "supports_prompt_caching": true, + "supports_system_messages": true, + "supports_response_schema": true }, "snowflake/deepseek-r1": { "litellm_provider": "snowflake", - "max_input_tokens": 32768, - "max_output_tokens": 8192, - "max_tokens": 8192, + "max_input_tokens": 128000, + "max_output_tokens": 16384, + "max_tokens": 16384, "mode": "chat", - "supports_reasoning": true + "input_cost_per_token": 0.00000135, + "output_cost_per_token": 0.0000054, + "supports_reasoning": true, + "supports_system_messages": true }, "snowflake/gemma-7b": { "litellm_provider": "snowflake", @@ -28147,23 +29731,34 @@ "snowflake/llama3.1-405b": { "litellm_provider": "snowflake", "max_input_tokens": 128000, - "max_output_tokens": 8192, - "max_tokens": 8192, - "mode": "chat" + "max_output_tokens": 16384, + "max_tokens": 16384, + "mode": "chat", + "input_cost_per_token": 0.0000012, + "output_cost_per_token": 0.0000012, + "supports_function_calling": true, + "supports_system_messages": true }, "snowflake/llama3.1-70b": { "litellm_provider": "snowflake", "max_input_tokens": 128000, - "max_output_tokens": 8192, - "max_tokens": 8192, - "mode": "chat" + "max_output_tokens": 16384, + "max_tokens": 16384, + "mode": "chat", + "input_cost_per_token": 7.2e-7, + "output_cost_per_token": 7.2e-7, + "supports_function_calling": true, + "supports_system_messages": true }, "snowflake/llama3.1-8b": { "litellm_provider": "snowflake", "max_input_tokens": 128000, - "max_output_tokens": 8192, - "max_tokens": 8192, - "mode": "chat" + "max_output_tokens": 16384, + "max_tokens": 16384, + "mode": "chat", + "input_cost_per_token": 2.4e-7, + "output_cost_per_token": 2.4e-7, + "supports_system_messages": true }, "snowflake/llama3.2-1b": { "litellm_provider": "snowflake", @@ -28180,11 +29775,26 @@ "mode": "chat" }, "snowflake/llama3.3-70b": { + "max_tokens": 16384, + "max_input_tokens": 128000, + "max_output_tokens": 16384, + "input_cost_per_token": 7.2e-7, + "output_cost_per_token": 7.2e-7, "litellm_provider": "snowflake", + "mode": "chat", + "supports_function_calling": true, + "supports_system_messages": true + }, + "snowflake/llama4-maverick": { + "max_tokens": 16384, "max_input_tokens": 128000, - "max_output_tokens": 8192, - "max_tokens": 8192, - "mode": "chat" + "max_output_tokens": 16384, + "input_cost_per_token": 2.4e-7, + "output_cost_per_token": 9.7e-7, + "litellm_provider": "snowflake", + "mode": "chat", + "supports_function_calling": true, + "supports_system_messages": true }, "snowflake/mistral-7b": { "litellm_provider": "snowflake", @@ -28203,9 +29813,14 @@ "snowflake/mistral-large2": { "litellm_provider": "snowflake", "max_input_tokens": 128000, - "max_output_tokens": 8192, - "max_tokens": 8192, - "mode": "chat" + "max_output_tokens": 16384, + "max_tokens": 16384, + "mode": "chat", + "input_cost_per_token": 0.000002, + "output_cost_per_token": 0.000006, + "supports_function_calling": true, + "supports_system_messages": true, + "supports_response_schema": true }, "snowflake/mixtral-8x7b": { "litellm_provider": "snowflake", @@ -28214,6 +29829,61 @@ "max_tokens": 8192, "mode": "chat" }, + "snowflake/openai-gpt-4.1": { + "max_tokens": 16384, + "max_input_tokens": 300000, + "max_output_tokens": 16384, + "input_cost_per_token": 0.000002, + "output_cost_per_token": 0.000008, + "cache_read_input_token_cost": 5e-7, + "litellm_provider": "snowflake", + "mode": "chat", + "supports_function_calling": true, + "supports_vision": true, + "supports_prompt_caching": true, + "supports_system_messages": true, + "supports_response_schema": true + }, + "snowflake/openai-gpt-5": { + "max_tokens": 16384, + "max_input_tokens": 300000, + "max_output_tokens": 16384, + "input_cost_per_token": 0.00000125, + "output_cost_per_token": 0.00001, + "cache_read_input_token_cost": 1.25e-7, + "litellm_provider": "snowflake", + "mode": "chat", + "supports_function_calling": true, + "supports_vision": true, + "supports_prompt_caching": true, + "supports_system_messages": true, + "supports_reasoning": true, + "supports_response_schema": true + }, + "snowflake/openai-gpt-5-mini": { + "max_tokens": 16384, + "max_input_tokens": 1000000, + "max_output_tokens": 16384, + "input_cost_per_token": 3e-7, + "output_cost_per_token": 0.0000012, + "litellm_provider": "snowflake", + "mode": "chat", + "supports_function_calling": true, + "supports_system_messages": true, + "supports_response_schema": true + }, + "snowflake/openai-gpt-5-nano": { + "max_tokens": 16384, + "max_input_tokens": 5000000, + "max_output_tokens": 16384, + "input_cost_per_token": 1.5e-7, + "output_cost_per_token": 6e-7, + "litellm_provider": "snowflake", + "mode": "chat", + "supports_function_calling": true, + "supports_system_messages": true, + "supports_response_schema": true + }, "snowflake/reka-core": { "litellm_provider": "snowflake", "max_input_tokens": 32000, @@ -28243,11 +29913,174 @@ "mode": "chat" }, "snowflake/snowflake-llama-3.3-70b": { + "max_tokens": 16384, + "max_input_tokens": 128000, + "max_output_tokens": 16384, + "input_cost_per_token": 7.2e-7, + "output_cost_per_token": 7.2e-7, "litellm_provider": "snowflake", - "max_input_tokens": 8000, - "max_output_tokens": 8192, - "max_tokens": 8192, - "mode": "chat" + "mode": "chat", + "supports_function_calling": true, + "supports_system_messages": true + }, + "tensormesh/MiniMaxAI/MiniMax-M2.5": { + "litellm_provider": "tensormesh", + "mode": "chat", + "input_cost_per_token": 3e-7, + "output_cost_per_token": 0.0000012, + "cache_read_input_token_cost": 0, + "max_input_tokens": 196608, + "max_output_tokens": 196608, + "supports_function_calling": true, + "supports_tool_choice": true, + "supports_response_schema": true, + "supports_prompt_caching": true, + "supports_system_messages": true, + "supports_reasoning": true, + "source": "https://serverless.tensormesh.ai/v1/models/openrouter" + }, + "tensormesh/Qwen/Qwen3-Coder-480B-A35B-Instruct-FP8": { + "litellm_provider": "tensormesh", + "mode": "chat", + "input_cost_per_token": 4.5e-7, + "output_cost_per_token": 0.0000018, + "cache_read_input_token_cost": 0, + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "supports_function_calling": true, + "supports_tool_choice": true, + "supports_response_schema": true, + "supports_prompt_caching": true, + "supports_system_messages": true, + "source": "https://serverless.tensormesh.ai/v1/models/openrouter" + }, + "tensormesh/Qwen/Qwen3.5-397B-A17B-FP8": { + "litellm_provider": "tensormesh", + "mode": "chat", + "input_cost_per_token": 6e-7, + "output_cost_per_token": 0.0000036, + "cache_read_input_token_cost": 0, + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "supports_function_calling": true, + "supports_tool_choice": true, + "supports_response_schema": true, + "supports_prompt_caching": true, + "supports_system_messages": true, + "supports_reasoning": true, + "source": "https://serverless.tensormesh.ai/v1/models/openrouter" + }, + "tensormesh/Qwen/Qwen3.6-27B-FP8": { + "litellm_provider": "tensormesh", + "mode": "chat", + "input_cost_per_token": 3.2e-7, + "output_cost_per_token": 0.0000032, + "cache_read_input_token_cost": 0, + "max_input_tokens": 262144, + "max_output_tokens": 262144, + "supports_function_calling": true, + "supports_tool_choice": true, + "supports_response_schema": true, + "supports_prompt_caching": true, + "supports_system_messages": true, + "supports_reasoning": true, + "source": "https://serverless.tensormesh.ai/v1/models/openrouter" + }, + "tensormesh/deepseek-ai/DeepSeek-V4-Flash": { + "litellm_provider": "tensormesh", + "mode": "chat", + "input_cost_per_token": 1.4e-7, + "output_cost_per_token": 2.8e-7, + "cache_read_input_token_cost": 0, + "max_input_tokens": 32768, + "max_output_tokens": 32768, + "supports_function_calling": true, + "supports_tool_choice": true, + "supports_response_schema": true, + "supports_prompt_caching": true, + "supports_system_messages": true, + "supports_reasoning": true, + "source": "https://serverless.tensormesh.ai/v1/models/openrouter" + }, + "tensormesh/google/gemma-4-31B-it": { + "litellm_provider": "tensormesh", + "mode": "chat", + "input_cost_per_token": 1.4e-7, + "output_cost_per_token": 5.6e-7, + "cache_read_input_token_cost": 0, + "max_input_tokens": 32768, + "max_output_tokens": 32768, + "supports_function_calling": true, + "supports_tool_choice": true, + "supports_response_schema": true, + "supports_prompt_caching": true, + "supports_system_messages": true, + "supports_reasoning": true, + "source": "https://serverless.tensormesh.ai/v1/models/openrouter" + }, + "tensormesh/lukealonso/GLM-5.1-NVFP4-MTP": { + "litellm_provider": "tensormesh", + "mode": "chat", + "input_cost_per_token": 0.0000014, + "output_cost_per_token": 0.0000044, + "cache_read_input_token_cost": 0, + "max_input_tokens": 202752, + "max_output_tokens": 202752, + "supports_function_calling": true, + "supports_tool_choice": true, + "supports_response_schema": true, + "supports_prompt_caching": true, + "supports_system_messages": true, + "supports_reasoning": true, + "source": "https://serverless.tensormesh.ai/v1/models/openrouter" + }, + "tensormesh/moonshotai/Kimi-K2.6": { + "litellm_provider": "tensormesh", + "mode": "chat", + "input_cost_per_token": 9.6e-7, + "output_cost_per_token": 0.000004, + "cache_read_input_token_cost": 0, + "max_input_tokens": 32768, + "max_output_tokens": 32768, + "supports_function_calling": true, + "supports_tool_choice": true, + "supports_response_schema": true, + "supports_prompt_caching": true, + "supports_system_messages": true, + "supports_reasoning": true, + "source": "https://serverless.tensormesh.ai/v1/models/openrouter" + }, + "tensormesh/openai/gpt-oss-120b": { + "litellm_provider": "tensormesh", + "mode": "chat", + "input_cost_per_token": 1.5e-7, + "output_cost_per_token": 6e-7, + "cache_read_input_token_cost": 0, + "max_input_tokens": 131072, + "max_output_tokens": 131072, + "supports_function_calling": true, + "supports_tool_choice": true, + "supports_response_schema": true, + "supports_prompt_caching": true, + "supports_system_messages": true, + "supports_reasoning": true, + "source": "https://serverless.tensormesh.ai/v1/models/openrouter" + }, + "tensormesh/openai/gpt-oss-20b": { + "litellm_provider": "tensormesh", + "mode": "chat", + "input_cost_per_token": 7e-8, + "output_cost_per_token": 2.8e-7, + "cache_read_input_token_cost": 0, + "max_input_tokens": 131072, + "max_output_tokens": 131072, + "supports_function_calling": true, + "supports_tool_choice": true, + "supports_response_schema": true, + "supports_prompt_caching": true, + "supports_system_messages": true, + "supports_reasoning": true, + "source": "https://serverless.tensormesh.ai/v1/models/openrouter" }, "together-ai-21.1b-41b": { "input_cost_per_token": 8e-7, @@ -28658,19 +30491,21 @@ "supports_video_input": true }, "us-gov.anthropic.claude-sonnet-4-5-20250929-v1:0": { - "cache_creation_input_token_cost": 0.000004125, - "cache_read_input_token_cost": 3.3e-7, - "input_cost_per_token": 0.0000033, - "input_cost_per_token_above_200k_tokens": 0.0000066, - "output_cost_per_token_above_200k_tokens": 0.00002475, - "cache_creation_input_token_cost_above_200k_tokens": 0.00000825, - "cache_read_input_token_cost_above_200k_tokens": 6.6e-7, + "cache_creation_input_token_cost": 0.0000045, + "cache_creation_input_token_cost_above_1hr": 0.0000072, + "cache_read_input_token_cost": 3.6e-7, + "input_cost_per_token": 0.0000036, + "input_cost_per_token_above_200k_tokens": 0.0000072, + "output_cost_per_token_above_200k_tokens": 0.000027, + "cache_creation_input_token_cost_above_200k_tokens": 0.000009, + "cache_creation_input_token_cost_above_1hr_above_200k_tokens": 0.0000144, + "cache_read_input_token_cost_above_200k_tokens": 7.2e-7, "litellm_provider": "bedrock_converse", "max_input_tokens": 200000, "max_output_tokens": 64000, "max_tokens": 64000, "mode": "chat", - "output_cost_per_token": 0.0000165, + "output_cost_per_token": 0.000018, "supports_assistant_prefill": true, "supports_computer_use": true, "supports_function_calling": true, @@ -28891,6 +30726,39 @@ "cache_read_input_token_cost": 3e-7, "cache_creation_input_token_cost": 0.00000375 }, + "us.anthropic.claude-fable-5": { + "cache_creation_input_token_cost": 0.00001375, + "cache_creation_input_token_cost_above_1hr": 0.000022, + "cache_read_input_token_cost": 0.0000011, + "input_cost_per_token": 0.000011, + "litellm_provider": "bedrock_converse", + "max_input_tokens": 1000000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "mode": "chat", + "output_cost_per_token": 0.000055, + "search_context_cost_per_query": { + "search_context_size_high": 0.01, + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01 + }, + "supports_adaptive_thinking": true, + "supports_assistant_prefill": false, + "supports_computer_use": true, + "supports_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_sampling_params": false, + "supports_tool_choice": true, + "supports_vision": true, + "supports_xhigh_reasoning_effort": true, + "supports_native_structured_output": true, + "supports_max_reasoning_effort": true, + "supports_output_config": true, + "bedrock_output_config_effort_ceiling": "xhigh" + }, "us.anthropic.claude-haiku-4-5-20251001-v1:0": { "cache_creation_input_token_cost": 0.000001375, "cache_creation_input_token_cost_above_1hr": 0.0000022, @@ -29046,6 +30914,7 @@ "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, + "supports_sampling_params": false, "supports_tool_choice": true, "supports_vision": true, "supports_xhigh_reasoning_effort": true, @@ -29070,6 +30939,7 @@ "search_context_size_low": 0.01, "search_context_size_medium": 0.01 }, + "supports_adaptive_thinking": true, "supports_assistant_prefill": false, "supports_computer_use": true, "supports_function_calling": true, @@ -29077,6 +30947,7 @@ "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, + "supports_sampling_params": false, "supports_tool_choice": true, "supports_vision": true, "supports_xhigh_reasoning_effort": true, @@ -29912,7 +31783,8 @@ "supports_vision": true, "supports_function_calling": true, "supports_tool_choice": true, - "supports_response_schema": true + "supports_response_schema": true, + "supports_image_size": false }, "vercel_ai_gateway/google/gemini-2.5-pro": { "input_cost_per_token": 0.0000025, @@ -30645,6 +32517,7 @@ }, "vertex_ai/claude-3-7-sonnet@20250219": { "cache_creation_input_token_cost": 0.00000375, + "cache_creation_input_token_cost_above_1hr": 0.000006, "cache_read_input_token_cost": 3e-7, "deprecation_date": "2026-05-11", "input_cost_per_token": 0.000003, @@ -30742,8 +32615,69 @@ "supports_tool_choice": true, "supports_vision": true }, + "vertex_ai/claude-fable-5": { + "cache_creation_input_token_cost": 0.0000125, + "cache_creation_input_token_cost_above_1hr": 0.00002, + "cache_read_input_token_cost": 0.000001, + "input_cost_per_token": 0.00001, + "litellm_provider": "vertex_ai-anthropic_models", + "max_input_tokens": 1000000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "mode": "chat", + "output_cost_per_token": 0.00005, + "search_context_cost_per_query": { + "search_context_size_high": 0.01, + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01 + }, + "supports_adaptive_thinking": true, + "supports_assistant_prefill": false, + "supports_computer_use": true, + "supports_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_sampling_params": false, + "supports_tool_choice": true, + "supports_vision": true, + "supports_xhigh_reasoning_effort": true, + "supports_max_reasoning_effort": true + }, + "vertex_ai/claude-fable-5@default": { + "cache_creation_input_token_cost": 0.0000125, + "cache_creation_input_token_cost_above_1hr": 0.00002, + "cache_read_input_token_cost": 0.000001, + "input_cost_per_token": 0.00001, + "litellm_provider": "vertex_ai-anthropic_models", + "max_input_tokens": 1000000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "mode": "chat", + "output_cost_per_token": 0.00005, + "search_context_cost_per_query": { + "search_context_size_high": 0.01, + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01 + }, + "supports_adaptive_thinking": true, + "supports_assistant_prefill": false, + "supports_computer_use": true, + "supports_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_sampling_params": false, + "supports_tool_choice": true, + "supports_vision": true, + "supports_xhigh_reasoning_effort": true, + "supports_max_reasoning_effort": true + }, "vertex_ai/claude-haiku-4-5": { "cache_creation_input_token_cost": 0.00000125, + "cache_creation_input_token_cost_above_1hr": 0.000002, "cache_read_input_token_cost": 1e-7, "input_cost_per_token": 0.000001, "litellm_provider": "vertex_ai-anthropic_models", @@ -30765,6 +32699,7 @@ }, "vertex_ai/claude-haiku-4-5@20251001": { "cache_creation_input_token_cost": 0.00000125, + "cache_creation_input_token_cost_above_1hr": 0.000002, "cache_read_input_token_cost": 1e-7, "input_cost_per_token": 0.000001, "litellm_provider": "vertex_ai-anthropic_models", @@ -30786,6 +32721,7 @@ }, "vertex_ai/claude-opus-4": { "cache_creation_input_token_cost": 0.00001875, + "cache_creation_input_token_cost_above_1hr": 0.00003, "cache_read_input_token_cost": 0.0000015, "input_cost_per_token": 0.000015, "litellm_provider": "vertex_ai-anthropic_models", @@ -30811,6 +32747,7 @@ }, "vertex_ai/claude-opus-4-1": { "cache_creation_input_token_cost": 0.00001875, + "cache_creation_input_token_cost_above_1hr": 0.00003, "cache_read_input_token_cost": 0.0000015, "input_cost_per_token": 0.000015, "input_cost_per_token_batches": 0.0000075, @@ -30828,6 +32765,7 @@ }, "vertex_ai/claude-opus-4-1@20250805": { "cache_creation_input_token_cost": 0.00001875, + "cache_creation_input_token_cost_above_1hr": 0.00003, "cache_read_input_token_cost": 0.0000015, "input_cost_per_token": 0.000015, "input_cost_per_token_batches": 0.0000075, @@ -30845,6 +32783,7 @@ }, "vertex_ai/claude-opus-4-5": { "cache_creation_input_token_cost": 0.00000625, + "cache_creation_input_token_cost_above_1hr": 0.00001, "cache_read_input_token_cost": 5e-7, "input_cost_per_token": 0.000005, "litellm_provider": "vertex_ai-anthropic_models", @@ -30871,6 +32810,7 @@ }, "vertex_ai/claude-opus-4-5@20251101": { "cache_creation_input_token_cost": 0.00000625, + "cache_creation_input_token_cost_above_1hr": 0.00001, "cache_read_input_token_cost": 5e-7, "input_cost_per_token": 0.000005, "litellm_provider": "vertex_ai-anthropic_models", @@ -30898,6 +32838,7 @@ }, "vertex_ai/claude-opus-4-6": { "cache_creation_input_token_cost": 0.00000625, + "cache_creation_input_token_cost_above_1hr": 0.00001, "cache_read_input_token_cost": 5e-7, "input_cost_per_token": 0.000005, "litellm_provider": "vertex_ai-anthropic_models", @@ -30925,6 +32866,7 @@ }, "vertex_ai/claude-opus-4-6@default": { "cache_creation_input_token_cost": 0.00000625, + "cache_creation_input_token_cost_above_1hr": 0.00001, "cache_read_input_token_cost": 5e-7, "input_cost_per_token": 0.000005, "litellm_provider": "vertex_ai-anthropic_models", @@ -30952,6 +32894,7 @@ }, "vertex_ai/claude-opus-4-7": { "cache_creation_input_token_cost": 0.00000625, + "cache_creation_input_token_cost_above_1hr": 0.00001, "cache_read_input_token_cost": 5e-7, "input_cost_per_token": 0.000005, "litellm_provider": "vertex_ai-anthropic_models", @@ -30972,6 +32915,7 @@ "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, + "supports_sampling_params": false, "supports_tool_choice": true, "supports_vision": true, "supports_xhigh_reasoning_effort": true, @@ -30979,6 +32923,7 @@ }, "vertex_ai/claude-opus-4-7@default": { "cache_creation_input_token_cost": 0.00000625, + "cache_creation_input_token_cost_above_1hr": 0.00001, "cache_read_input_token_cost": 5e-7, "input_cost_per_token": 0.000005, "litellm_provider": "vertex_ai-anthropic_models", @@ -30999,6 +32944,7 @@ "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, + "supports_sampling_params": false, "supports_tool_choice": true, "supports_vision": true, "supports_xhigh_reasoning_effort": true, @@ -31020,6 +32966,7 @@ "search_context_size_low": 0.01, "search_context_size_medium": 0.01 }, + "supports_adaptive_thinking": true, "supports_assistant_prefill": false, "supports_computer_use": true, "supports_function_calling": true, @@ -31027,6 +32974,7 @@ "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, + "supports_sampling_params": false, "supports_tool_choice": true, "supports_vision": true, "supports_xhigh_reasoning_effort": true, @@ -31048,6 +32996,7 @@ "search_context_size_low": 0.01, "search_context_size_medium": 0.01 }, + "supports_adaptive_thinking": true, "supports_assistant_prefill": false, "supports_computer_use": true, "supports_function_calling": true, @@ -31055,6 +33004,7 @@ "supports_prompt_caching": true, "supports_reasoning": true, "supports_response_schema": true, + "supports_sampling_params": false, "supports_tool_choice": true, "supports_vision": true, "supports_xhigh_reasoning_effort": true, @@ -31062,6 +33012,7 @@ }, "vertex_ai/claude-opus-4@20250514": { "cache_creation_input_token_cost": 0.00001875, + "cache_creation_input_token_cost_above_1hr": 0.00003, "cache_read_input_token_cost": 0.0000015, "input_cost_per_token": 0.000015, "litellm_provider": "vertex_ai-anthropic_models", @@ -31087,6 +33038,7 @@ }, "vertex_ai/claude-sonnet-4": { "cache_creation_input_token_cost": 0.00000375, + "cache_creation_input_token_cost_above_1hr": 0.000006, "cache_read_input_token_cost": 3e-7, "input_cost_per_token": 0.000003, "input_cost_per_token_above_200k_tokens": 0.000006, @@ -31116,6 +33068,7 @@ }, "vertex_ai/claude-sonnet-4-5": { "cache_creation_input_token_cost": 0.00000375, + "cache_creation_input_token_cost_above_1hr": 0.000006, "cache_read_input_token_cost": 3e-7, "input_cost_per_token": 0.000003, "input_cost_per_token_above_200k_tokens": 0.000006, @@ -31142,6 +33095,7 @@ }, "vertex_ai/claude-sonnet-4-5@20250929": { "cache_creation_input_token_cost": 0.00000375, + "cache_creation_input_token_cost_above_1hr": 0.000006, "cache_read_input_token_cost": 3e-7, "input_cost_per_token": 0.000003, "input_cost_per_token_above_200k_tokens": 0.000006, @@ -31169,6 +33123,7 @@ }, "vertex_ai/claude-sonnet-4-6": { "cache_creation_input_token_cost": 0.00000375, + "cache_creation_input_token_cost_above_1hr": 0.000006, "cache_read_input_token_cost": 3e-7, "input_cost_per_token": 0.000003, "litellm_provider": "vertex_ai-anthropic_models", @@ -31196,6 +33151,7 @@ }, "vertex_ai/claude-sonnet-4-6@default": { "cache_creation_input_token_cost": 0.00000375, + "cache_creation_input_token_cost_above_1hr": 0.000006, "cache_read_input_token_cost": 3e-7, "input_cost_per_token": 0.000003, "litellm_provider": "vertex_ai-anthropic_models", @@ -31223,6 +33179,7 @@ }, "vertex_ai/claude-sonnet-4@20250514": { "cache_creation_input_token_cost": 0.00000375, + "cache_creation_input_token_cost_above_1hr": 0.000006, "cache_read_input_token_cost": 3e-7, "input_cost_per_token": 0.000003, "input_cost_per_token_above_200k_tokens": 0.000006, @@ -31408,7 +33365,8 @@ "supports_url_context": true, "supports_vision": true, "supports_web_search": false, - "tpm": 8000000 + "tpm": 8000000, + "supports_image_size": false }, "vertex_ai/gemini-3-flash-preview": { "cache_read_input_token_cost": 5e-8, @@ -31837,6 +33795,22 @@ }, "web_search_billing_unit": "per_query" }, + "vertex_ai/google/gemma-4-26b-a4b-it-maas": { + "input_cost_per_token": 1.5e-7, + "litellm_provider": "vertex_ai-openai_models", + "max_input_tokens": 256000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "mode": "chat", + "output_cost_per_token": 6e-7, + "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/maas/google/gemma-4-26b-a4b-it", + "supported_regions": [ + "global" + ], + "supports_function_calling": true, + "supports_tool_choice": true, + "supports_vision": true + }, "vertex_ai/jamba-1.5": { "input_cost_per_token": 2e-7, "litellm_provider": "vertex_ai-ai21_models", @@ -33190,7 +35164,8 @@ "supports_prompt_caching": true, "supports_response_schema": false, "supports_tool_choice": true, - "supports_web_search": true + "supports_web_search": true, + "deprecation_date": "2026-05-15" }, "xai/grok-3-beta": { "cache_read_input_token_cost": 7.5e-7, @@ -33388,7 +35363,8 @@ "supports_function_calling": true, "supports_prompt_caching": true, "supports_tool_choice": true, - "supports_web_search": true + "supports_web_search": true, + "deprecation_date": "2026-05-15" }, "xai/grok-4-1-fast": { "cache_read_input_token_cost": 5e-8, @@ -33429,7 +35405,8 @@ "supports_response_schema": true, "supports_tool_choice": true, "supports_vision": true, - "supports_web_search": true + "supports_web_search": true, + "deprecation_date": "2026-05-15" }, "xai/grok-4-1-fast-non-reasoning-latest": { "cache_read_input_token_cost": 5e-8, @@ -33449,7 +35426,8 @@ "supports_response_schema": true, "supports_tool_choice": true, "supports_vision": true, - "supports_web_search": true + "supports_web_search": true, + "deprecation_date": "2026-05-15" }, "xai/grok-4-1-fast-reasoning": { "cache_read_input_token_cost": 5e-8, @@ -33470,7 +35448,8 @@ "supports_response_schema": true, "supports_tool_choice": true, "supports_vision": true, - "supports_web_search": true + "supports_web_search": true, + "deprecation_date": "2026-05-15" }, "xai/grok-4-1-fast-reasoning-latest": { "cache_read_input_token_cost": 5e-8, @@ -33491,7 +35470,8 @@ "supports_response_schema": true, "supports_tool_choice": true, "supports_vision": true, - "supports_web_search": true + "supports_web_search": true, + "deprecation_date": "2026-05-15" }, "xai/grok-4-fast-non-reasoning": { "cache_read_input_token_cost": 5e-8, @@ -33508,7 +35488,8 @@ "supports_function_calling": true, "supports_prompt_caching": true, "supports_tool_choice": true, - "supports_web_search": true + "supports_web_search": true, + "deprecation_date": "2026-05-15" }, "xai/grok-4-fast-reasoning": { "cache_read_input_token_cost": 5e-8, @@ -33525,7 +35506,8 @@ "supports_function_calling": true, "supports_prompt_caching": true, "supports_tool_choice": true, - "supports_web_search": true + "supports_web_search": true, + "deprecation_date": "2026-05-15" }, "xai/grok-4-latest": { "input_cost_per_token": 0.000003, @@ -33692,7 +35674,8 @@ "supports_function_calling": true, "supports_prompt_caching": true, "supports_reasoning": true, - "supports_tool_choice": true + "supports_tool_choice": true, + "deprecation_date": "2026-05-15" }, "xai/grok-code-fast-1-0825": { "cache_read_input_token_cost": 2e-8, @@ -33707,7 +35690,8 @@ "supports_function_calling": true, "supports_prompt_caching": true, "supports_reasoning": true, - "supports_tool_choice": true + "supports_tool_choice": true, + "deprecation_date": "2026-05-15" }, "xai/grok-vision-beta": { "input_cost_per_image": 0.000005, diff --git a/cecli/scrape.py b/cecli/scrape.py index 0722e11768d..390e3e9b295 100755 --- a/cecli/scrape.py +++ b/cecli/scrape.py @@ -220,7 +220,7 @@ def scrape_with_httpx(self, url): return None, None def try_pandoc(self): - if self.pandoc_available: + if self.pandoc_available is not None: return try: @@ -232,11 +232,9 @@ def try_pandoc(self): try: pypandoc.download_pandoc(delete_installer=True) - except Exception as err: - self.print_error(f"Unable to install pandoc: {err}") - return - - self.pandoc_available = True + self.pandoc_available = True + except Exception: + self.pandoc_available = False def html_to_markdown(self, page_source): from bs4 import BeautifulSoup diff --git a/cecli/tools/__init__.py b/cecli/tools/__init__.py index 44e527cff37..4791865349d 100644 --- a/cecli/tools/__init__.py +++ b/cecli/tools/__init__.py @@ -6,7 +6,6 @@ _yield, command, command_interactive, - context_manager, delegate, edit_text, explore_code, @@ -17,10 +16,9 @@ git_show, git_status, grep, - load_skill, ls, read_range, - remove_skill, + resource_manager, thinking, undo_change, update_todo_list, @@ -30,7 +28,6 @@ TOOL_MODULES = [ command, command_interactive, - context_manager, delegate, edit_text, explore_code, @@ -42,10 +39,9 @@ git_show, git_status, grep, - load_skill, ls, read_range, - remove_skill, + resource_manager, thinking, undo_change, update_todo_list, diff --git a/cecli/tools/command.py b/cecli/tools/command.py index f09d32f2cdf..646fd46d5ba 100644 --- a/cecli/tools/command.py +++ b/cecli/tools/command.py @@ -3,6 +3,8 @@ import os import platform +import xxhash + from cecli.helpers.background_commands import BackgroundCommandManager from cecli.run_cmd import run_cmd_subprocess from cecli.tools.utils.base_tool import BaseTool @@ -14,6 +16,7 @@ class Tool(BaseTool): NORM_NAME = "command" TRACK_INVOCATIONS = False + ALLOWED_SESSION_COMMANDS = {} SCHEMA = { "type": "function", "function": { @@ -70,6 +73,14 @@ def _parse_command_key(command): return command.split("::", 1)[1].strip() return None + @staticmethod + def _hash_command(command): + """Compute an xxhash of the full command text for session tracking.""" + if not command: + return command + + return xxhash.xxh64(command).hexdigest() + @classmethod async def execute( cls, coder, command, background=False, stop=None, stdin=None, pty=False, **kwargs @@ -124,6 +135,8 @@ async def execute( if not confirmed: return "Command execution skipped by user." + command = coder.format_command_with_prefix(command) + # Determine timeout from agent_config (default: 30 seconds) timeout = 0 if hasattr(coder, "agent_config"): @@ -139,6 +152,15 @@ async def execute( @classmethod async def _get_confirmation(cls, coder, command_string, background): """Get user confirmation for command execution.""" + # Hash command for dict key lookup + command_hash = cls._hash_command(command_string) + + # Check if command is already handled for this session + if command_hash in cls.ALLOWED_SESSION_COMMANDS: + if cls.ALLOWED_SESSION_COMMANDS[command_hash]: + return True # Previously approved for session + # Previously declined - skip session question, continue to normal confirmation + if coder.skip_cli_confirmations: return True @@ -150,21 +172,32 @@ async def _get_confirmation(cls, coder, command_string, background): if fnmatch.fnmatch(command_string, pattern): return True - command_string = coder.format_command_with_prefix(command_string) + formatted_command = coder.format_command_with_prefix(command_string) if background: prompt = "Allow execution of this background command?" else: prompt = "Allow execution of this command?" - return await coder.io.confirm_ask( + confirmed = await coder.io.confirm_ask( prompt, - subject=command_string, + subject=formatted_command, explicit_yes_required=True, allow_never=True, group_response="Command Tool", ) + if confirmed: + # Ask if user wants to allow for the entire session (only once per command) + if command_hash not in cls.ALLOWED_SESSION_COMMANDS: + session_allowed = await coder.io.confirm_ask( + "Allow this command for the rest of the session?", + subject=formatted_command, + ) + cls.ALLOWED_SESSION_COMMANDS[command_hash] = session_allowed + + return confirmed + @classmethod async def _execute_background(cls, coder, command_string, use_pty=False): """ @@ -277,8 +310,8 @@ async def _execute_with_timeout(cls, coder, command_string, timeout, use_pty=Fal output_content = ( f"[Large Response ({total_size} characters). " "Output saved to paginated files.]\n" - f"File Aliases (for use with ContextManager):\n{alias_list_str}\n" - "Use the `ContextManager` tool to view these files." + f"File Aliases (for use with ResourceManager):\n{alias_list_str}\n" + "Use the `ResourceManager` tool to view these files." "Do not use standard cli tools to view these files." "Remove them from context after taking notes on the relevant information " "to prevent overfilling stale context." @@ -364,8 +397,8 @@ async def _execute_foreground(cls, coder, command_string): output_content = ( f"[Large Response ({total_size} characters). " "Output saved to paginated files.]\n" - f"File Aliases (for use with ContextManager):\n{alias_list_str}\n" - "Use the `ContextManager` tool to view these files." + f"File Aliases (for use with ResourceManager):\n{alias_list_str}\n" + "Use the `ResourceManager` tool to view these files." "Do not use standard cli tools to view these files." "Remove them from context after taking note of the relevant information " "in the output to prevent overfilling stale context." diff --git a/cecli/tools/command_interactive.py b/cecli/tools/command_interactive.py index 30a27a60ec8..1a24a66093d 100644 --- a/cecli/tools/command_interactive.py +++ b/cecli/tools/command_interactive.py @@ -2,6 +2,8 @@ import asyncio import fnmatch +import xxhash + from cecli.run_cmd import run_cmd from cecli.tools.utils.base_tool import BaseTool @@ -9,6 +11,7 @@ class Tool(BaseTool): NORM_NAME = "commandinteractive" TRACK_INVOCATIONS = False + ALLOWED_SESSION_COMMANDS = {} SCHEMA = { "type": "function", "function": { @@ -36,40 +39,74 @@ def _is_command_allowed(coder, command_string): """Check if command matches any allowed_commands patterns.""" if hasattr(coder, "agent_config"): allowed_commands = coder.agent_config.get("allowed_commands", []) - for pattern in allowed_commands: - if fnmatch.fnmatch(command_string, pattern): - return True - + if allowed_commands: + for pattern in allowed_commands: + if fnmatch.fnmatch(command_string, pattern): + return True return False + @staticmethod + def _hash_command(command): + """Compute an xxhash of the full command text for session tracking.""" + if not command: + return command + + return xxhash.xxh64(command).hexdigest() + + @classmethod + async def _get_confirmation(cls, coder, command_string): + """Get user confirmation for command execution.""" + # Hash command for dict key lookup + command_hash = cls._hash_command(command_string) + + # Check if command is already handled for this session + if command_hash in cls.ALLOWED_SESSION_COMMANDS: + if cls.ALLOWED_SESSION_COMMANDS[command_hash]: + return True # Previously approved for session + # Previously declined - skip session question, continue to normal confirmation + + if coder.skip_cli_confirmations: + return True + + # Check if command matches any allowed_commands patterns + if cls._is_command_allowed(coder, command_string): + return True + + formatted_command = coder.format_command_with_prefix(command_string) + + confirmed = await coder.io.confirm_ask( + "Allow execution of this command?", + subject=formatted_command, + explicit_yes_required=True, + allow_never=True, + group_response="Command Interactive Tool", + ) + + if not confirmed: + return False + + # Ask if user wants to allow for the entire session (only once per command) + if command_hash not in cls.ALLOWED_SESSION_COMMANDS: + session_allowed = await coder.io.confirm_ask( + "Allow this command for the rest of the session?", + subject=formatted_command, + ) + cls.ALLOWED_SESSION_COMMANDS[command_hash] = session_allowed + + return True + @classmethod async def execute(cls, coder, command_string, **kwargs): """ Execute an interactive shell command using run_cmd (which uses pexpect/PTY). """ try: - command_string = coder.format_command_with_prefix(command_string) - - confirmed = ( - True - if coder.skip_cli_confirmations or cls._is_command_allowed(coder, command_string) - else await coder.io.confirm_ask( - "Allow execution of this command?", - subject=command_string, - explicit_yes_required=True, # Require explicit 'yes' or 'always' - allow_never=True, # Enable the 'Always' option - group_response="Command Interactive Tool", - ) - ) - + confirmed = await cls._get_confirmation(coder, command_string) if not confirmed: - # This happens if the user explicitly says 'no' this time. - # If 'Always' was chosen previously, confirm_ask returns True directly. - coder.io.tool_output( - f"Skipped execution of shell command: {command_string}", type="tool-result" - ) return "Shell command execution skipped by user." + command_string = coder.format_command_with_prefix(command_string) + coder.io.tool_output( f"⛭ Starting interactive shell command: {command_string}", type="tool-result" ) @@ -86,7 +123,6 @@ def _run_interactive(): ) if tui: - # Notify user and suspend TUI for interactive command coder.io.tool_output( ">>> Suspending TUI for interactive command <<<", type="tool-result" ) @@ -107,10 +143,8 @@ def _run_interactive(): # Format the output for the result message, include more content output_content = combined_output or "" - # Use the existing token threshold constant as the character limit for truncation output_limit = coder.large_file_token_threshold if coder.context_management_enabled and len(output_content) > output_limit: - # Truncate and add a clear message using the constant value output_content = ( output_content[:output_limit] + f"\n... (output truncated at {output_limit} characters, based on" diff --git a/cecli/tools/context_manager.py b/cecli/tools/context_manager.py deleted file mode 100644 index c5883f97e16..00000000000 --- a/cecli/tools/context_manager.py +++ /dev/null @@ -1,308 +0,0 @@ -import os -import re -import time - -from cecli.helpers.background_commands import BackgroundCommandManager -from cecli.tools.utils.base_tool import BaseTool -from cecli.tools.utils.helpers import ToolError, parse_arg_as_list -from cecli.tools.utils.output import color_markers, tool_footer, tool_header -from cecli.tools.validations import ToolValidations - - -class Tool(BaseTool): - NORM_NAME = "contextmanager" - SCHEMA = { - "type": "function", - "function": { - "name": "ContextManager", - "description": ( - "Manage multiple files in the chat context: add, read_only, create, and remove." - " Accepts arrays of file paths for each operation." - ), - "parameters": { - "type": "object", - "properties": { - "add": { - "type": "array", - "items": {"type": "string"}, - "description": ( - "List of file paths to add to context. Limit to at most 2 at a time." - ), - }, - "read_only": { - "type": "array", - "items": {"type": "string"}, - "description": ( - "List of file paths to add as read-only. Limit to at most 2 at a time." - ), - }, - "create": { - "type": "array", - "items": {"type": "string"}, - "description": "List of file paths to create.", - }, - "remove": { - "type": "array", - "items": {"type": "string"}, - "description": "List of file paths to remove from context.", - }, - "stop": { - "type": "array", - "items": {"type": "string"}, - "description": "List of command keys to stop background commands for.", - }, - }, - "additionalProperties": False, - "required": [], - }, - }, - } - - @classmethod - def execute( - cls, coder, remove=None, add=None, read_only=None, create=None, stop=None, **kwargs - ): - """Perform batch operations on the coder's context. - - Parameters - ---------- - coder: Coder instance - The active coder handling file context. - remove: list[str] | None - Files to remove from the context. - add: list[str] | None - Files to promote to editable status. - view: list[str] | None - Files to add as read-only view. - create: list[str] | None - Files to create and make editable. - """ - remove_files = sorted(parse_arg_as_list(remove), key=cls._natural_sort_key) - editable_files = sorted(parse_arg_as_list(add), key=cls._natural_sort_key) - view_files = sorted(parse_arg_as_list(read_only), key=cls._natural_sort_key) - create_files = sorted(parse_arg_as_list(create), key=cls._natural_sort_key) - stop_keys = sorted(parse_arg_as_list(stop), key=cls._natural_sort_key) - - if ( - not remove_files - and not editable_files - and not view_files - and not create_files - and not stop_keys - ): - raise ToolError( - "You must specify at least one of: remove, editable, view, create, or stop" - ) - - coder.io.tool_output("⛭ Modifying Context", type="tool-result") - messages = [] - - for f in create_files: - messages.append(cls._create(coder, f)) - for f in remove_files: - messages.append(cls._remove(coder, f)) - for f in view_files: - messages.append(cls._view(coder, f)) - for f in editable_files: - messages.append(cls._editable(coder, f)) - for key in stop_keys: - messages.append(cls._stop_command(coder, key)) - - if coder.tui and coder.tui(): - coder.tui().refresh() - - coder.context_blocks_cache = {} - coder.edit_allowed = True - - return "\n".join(messages) - - @classmethod - def format_output(cls, coder, mcp_server, tool_response): - """Format output for ContextManager tool.""" - color_start, color_end = color_markers(coder) - - # Output header - tool_header(coder=coder, mcp_server=mcp_server, tool_response=tool_response) - - try: - params = ToolValidations.validate_params( - tool_response.function.arguments, cls.VALIDATIONS, cls.SCHEMA - ) - except ToolError: - coder.io.tool_error("Invalid Tool JSON") - return - - # Define action display names - action_names = { - "create": "create", - "remove": "remove", - "view": "view", - "editable": "editable", - "stop": "stop", - } - - # Output each action with comma-separated file list - for action_key, display_name in action_names.items(): - files = sorted(parse_arg_as_list(params.get(action_key)), key=cls._natural_sort_key) - if files: - file_list = ", ".join(files) - coder.io.tool_output(f"{color_start}{display_name}:{color_end} {file_list}") - - tool_footer(coder=coder, tool_response=tool_response, params=params) - - @classmethod - def _remove(cls, coder, file_path): - """Remove a file from the coder's context.""" - from cecli.helpers.conversation import ConversationService - - try: - abs_path = cls._resolve_file_path(coder, file_path) - rel_path = coder.get_rel_fname(abs_path) - removed = False - - if abs_path in coder.abs_fnames: - coder.abs_fnames.remove(abs_path) - removed = True - - if abs_path in coder.abs_read_only_fnames: - coder.abs_read_only_fnames.remove(abs_path) - removed = True - - if not removed: - coder.io.tool_output(f"⚠ File '{file_path}' not in context", type="tool-result") - return f"File not in context: {file_path}" - - coder.recently_removed[rel_path] = {"removed_at": time.time()} - - if not file_path.startswith("command_key::"): - ConversationService.get_chunks(coder).defer_removal(abs_path) - ConversationService.get_chunks(coder).defer_removal(rel_path) - - coder.io.tool_output(f"✗ Removed '{file_path}' from context", type="tool-result") - return ( - f"Removed: {file_path}\n" - "Old file contents may remain visible. This is an acceptable system behavior." - ) - except Exception as e: - coder.io.tool_error(f"Error removing file '{file_path}': {str(e)}") - return f"Error removing {file_path}: {e}" - - @classmethod - def _stop_command(cls, coder, command_key): - """Stop a background command by its command key.""" - try: - success, output, exit_code = BackgroundCommandManager.stop_background_command( - command_key - ) - if success: - coder.io.tool_output( - f"✗ Stopped background command '{command_key}'", type="tool-result" - ) - return ( - f"Background command stopped: {command_key}\n" - f"Exit code: {exit_code}\n" - f"Final output:\n{output}" - ) - else: - coder.io.tool_output( - f"⚠ Background command '{command_key}' not found or not running", - type="tool-result", - ) - return f"Command not found or not running: {command_key}" - except Exception as e: - coder.io.tool_error(f"Error stopping command '{command_key}': {str(e)}") - return f"Error stopping {command_key}: {e}" - - @classmethod - def _editable(cls, coder, file_path): - """Make a file editable in the coder's context.""" - try: - abs_path = cls._resolve_file_path(coder, file_path) - if abs_path in coder.abs_fnames: - coder.io.tool_output( - f"🗀 File '{file_path}' is already editable", type="tool-result" - ) - return f"Already editable: {file_path}" - if not os.path.isfile(abs_path): - coder.io.tool_output(f"⚠ File '{file_path}' not found on disk", type="tool-result") - return f"File not found: {file_path}" - was_read_only = False - if abs_path in coder.abs_read_only_fnames: - coder.abs_read_only_fnames.remove(abs_path) - was_read_only = True - coder.abs_fnames.add(abs_path) - if was_read_only: - coder.io.tool_output( - f"🗀 Moved '{file_path}' from read-only to editable", type="tool-result" - ) - return f"Made editable (moved): {file_path}" - else: - coder.io.tool_output( - f"🗀 Added '{file_path}' directly to editable context", type="tool-result" - ) - return f"Made editable (added): {file_path}" - except Exception as e: - coder.io.tool_error(f"Error making editable '{file_path}': {str(e)}") - return f"Error making editable {file_path}: {e}" - - @classmethod - def _view(cls, coder, file_path): - """View a file (add as read‑only) in the coder's context.""" - try: - resolved_path = cls._resolve_file_path(coder, file_path) - return coder._add_file_to_context(resolved_path, explicit=True) - except Exception as e: - coder.io.tool_error(f"Error viewing file '{file_path}': {str(e)}") - return f"Error viewing {file_path}: {e}" - - @classmethod - def _create(cls, coder, file_path): - """Create a new file on the file system and make it editable in the coder's context.""" - try: - abs_path = coder.abs_root_path(file_path) - - # Check if file already exists - if os.path.exists(abs_path): - coder.io.tool_output(f"⚠ File '{file_path}' already exists", type="tool-result") - return f"File already exists: {file_path}" - - # Create parent directories if they don't exist - os.makedirs(os.path.dirname(abs_path), exist_ok=True) - - # Create an empty file - with open(abs_path, "w", encoding="utf-8"): - pass - - # Add the file to editable context - coder.abs_fnames.add(abs_path) - - coder.io.tool_output( - f"🗀 Created '{file_path}' and made it editable", type="tool-result" - ) - return f"Created and made editable: {file_path}" - - except Exception as e: - coder.io.tool_error(f"Error creating file '{file_path}': {str(e)}") - return f"Error creating {file_path}: {e}" - - @classmethod - def _resolve_file_path(cls, coder, file_path): - """Resolve a file path, handling command_key:: aliases. - - command_key::{command_key}/{filename} resolves to the actual - file path under the agent's local agent folder. - """ - if file_path.startswith("command_key::"): - alias_path = file_path[len("command_key::") :] - parts = alias_path.split("/", 1) - if len(parts) == 2: - command_key = parts[0] - filename = parts[1] - rel_path = coder.local_agent_folder(f"{command_key}/{filename}") - return coder.abs_root_path(rel_path) - return coder.abs_root_path(file_path) - - @classmethod - def _natural_sort_key(cls, s: str) -> list: - """Natural sort key that splits "a10b2" into ["a", 10, "b", 2].""" - return [int(text) if text.isdigit() else text.lower() for text in re.split(r"(\d+)", s)] diff --git a/cecli/tools/edit_text.py b/cecli/tools/edit_text.py index a22e96dcd57..15dfb18730f 100644 --- a/cecli/tools/edit_text.py +++ b/cecli/tools/edit_text.py @@ -88,7 +88,13 @@ class Tool(BaseTool): ), }, }, - "required": ["file_path"], + "required": [ + "file_path", + "operation", + "start_line", + "end_line", + "text", + ], }, "description": "Array of edits to apply.", }, @@ -181,12 +187,18 @@ def execute( ) edit_text_raw = edit.get("text") - edit_text = ( - strip_hashline(edit_text_raw) if edit_text_raw is not None else None - ) + edit_text = edit.get("text") edit_start_line = edit.get("start_line") edit_end_line = edit.get("end_line") + if edit_text_raw is not None: + edit_text_raw = strip_hashline(edit_text_raw) + while edit_text_raw != edit_text: + edit_text_raw = strip_hashline(edit_text_raw) + edit_text = strip_hashline(edit_text) + + edit_text = edit_text_raw + # Try to resolve line content values to content IDs # This handles cases where LLMs pass actual line content # instead of content ID markers @@ -437,7 +449,7 @@ def format_output(cls, coder, mcp_server, tool_response): start_line = edit.get("start_line") end_line = edit.get("end_line") # Show output based on operation type - if operation == "replace": + if operation in ("replace", "delete"): # Show diff for replace operations diff_output = "" @@ -451,7 +463,7 @@ def format_output(cls, coder, mcp_server, tool_response): original_content=strip_hashline(original_content), start_line_hash=start_line, end_line_hash=end_line, - operation="replace", + operation=operation, text=strip_hashline(text), ) except ContentHashError as e: @@ -469,14 +481,4 @@ def format_output(cls, coder, mcp_server, tool_response): coder.io.tool_output(text) coder.io.tool_output("") - elif operation == "delete": - # Show deletion summary - range_info = ( - f"Deleted {start_line} - {end_line}" - if start_line and end_line - else "specified range" - ) - coder.io.tool_output(range_info) - coder.io.tool_output("") - tool_footer(coder=coder, tool_response=tool_response, params=params) diff --git a/cecli/tools/load_skill.py b/cecli/tools/load_skill.py deleted file mode 100644 index f59beea940f..00000000000 --- a/cecli/tools/load_skill.py +++ /dev/null @@ -1,48 +0,0 @@ -from cecli.tools.utils.base_tool import BaseTool - - -class Tool(BaseTool): - NORM_NAME = "loadskill" - SCHEMA = { - "type": "function", - "function": { - "name": "LoadSkill", - "description": "Load a skill by name.", - "parameters": { - "type": "object", - "properties": { - "skill_name": { - "type": "string", - "description": "Name of the skill to load", - }, - }, - "required": ["skill_name"], - }, - }, - } - - @classmethod - def execute(cls, coder, skill_name, **kwargs): - """ - Load a skill by name (agent mode only). - """ - if not skill_name: - return "Error: Skill name is required." - - # Check if we're in agent mode - if not hasattr(coder, "edit_format") or coder.edit_format not in ("agent", "subagent"): - return "Error: Skill loading is only available in agent mode." - - # Check if skills_manager is available - if not hasattr(coder, "skills_manager") or coder.skills_manager is None: - error_msg = "Error: Skills manager is not initialized. Skills may not be configured." - # Check if skills directories are configured - if hasattr(coder, "skills_directory_paths") and not coder.skills_directory_paths: - error_msg += ( - "\nNo skills directories configured. Use --skills-paths to configure skill" - " directories." - ) - return error_msg - - # Use the instance method on skills_manager - return coder.skills_manager.load_skill(skill_name) diff --git a/cecli/tools/read_range.py b/cecli/tools/read_range.py index cba2bfd6f29..67a46090c4b 100644 --- a/cecli/tools/read_range.py +++ b/cecli/tools/read_range.py @@ -212,7 +212,7 @@ def execute(cls, coder, read, **kwargs): f"File {rel_path} is empty.", ( "Next: use EditText with start_line @000 and end_line @000 to" - " write content, or ContextManager to scaffold — do not call" + " write content, or ResourceManager to scaffold — do not call" " ReadRange again on this empty file." ), ] @@ -711,7 +711,7 @@ def format_model_response(cls, coder, rel_path, s_idx, e_idx, hashed_lines, curr if start_found or end_found: if start_found: lines.append( - f"File {rel_path} Snapshot (Lines {start_stub_s + 1} - {start_stub_e + 1}):" + f"File {rel_path} Current Snapshot (Lines {start_stub_s + 1} - {start_stub_e + 1}):" ) lines.extend(hashed_lines[start_stub_s:start_stub_e]) @@ -723,7 +723,7 @@ def format_model_response(cls, coder, rel_path, s_idx, e_idx, hashed_lines, curr ): lines.append("...⋮...") lines.append( - f"File {rel_path} Snapshot (Lines {end_stub_s + 1} - {end_stub_e + 1}):" + f"File {rel_path} Current Snapshot (Lines {end_stub_s + 1} - {end_stub_e + 1}):" ) lines.extend(hashed_lines[end_stub_s:end_stub_e]) @@ -732,14 +732,21 @@ def format_model_response(cls, coder, rel_path, s_idx, e_idx, hashed_lines, curr except Exception: pass - lines = [f"File {rel_path} Snapshot (Lines {s_idx + 1} - {e_idx + 1}):"] + lines = [f"File {rel_path} Current Snapshot (Lines {s_idx + 1} - {e_idx + 1}):"] total = e_idx - s_idx - if total <= 15: + hashed_content = "\n".join(hashed_lines[s_idx : e_idx + 1]) + token_count = coder.main_model.token_count(hashed_content) + + if token_count <= min(coder.large_file_token_threshold / 16, 512): lines.extend(hashed_lines[s_idx : e_idx + 1]) else: - lines.extend(hashed_lines[s_idx : s_idx + 5]) - lines.append("...⋮...") - lines.extend(hashed_lines[e_idx - 4 : e_idx + 1]) + if total <= 15: + lines.extend(hashed_lines[s_idx : e_idx + 1]) + else: + lines.extend(hashed_lines[s_idx : s_idx + 5]) + lines.append("...⋮...") + lines.extend(hashed_lines[e_idx - 4 : e_idx + 1]) + lines.append("") return "\n".join(lines) diff --git a/cecli/tools/remove_skill.py b/cecli/tools/remove_skill.py deleted file mode 100644 index a4e0e72eed9..00000000000 --- a/cecli/tools/remove_skill.py +++ /dev/null @@ -1,48 +0,0 @@ -from cecli.tools.utils.base_tool import BaseTool - - -class Tool(BaseTool): - NORM_NAME = "removeskill" - SCHEMA = { - "type": "function", - "function": { - "name": "RemoveSkill", - "description": "Remove a skill by name.", - "parameters": { - "type": "object", - "properties": { - "skill_name": { - "type": "string", - "description": "Name of the skill to remove", - }, - }, - "required": ["skill_name"], - }, - }, - } - - @classmethod - def execute(cls, coder, skill_name, **kwargs): - """ - Remove a skill by name (agent mode only). - """ - if not skill_name: - return "Error: Skill name is required." - - # Check if we're in agent mode - if not hasattr(coder, "edit_format") or coder.edit_format not in ("agent", "subagent"): - return "Error: Skill removal is only available in agent mode." - - # Check if skills_manager is available - if not hasattr(coder, "skills_manager") or coder.skills_manager is None: - error_msg = "Error: Skills manager is not initialized. Skills may not be configured." - # Check if skills directories are configured - if hasattr(coder, "skills_directory_paths") and not coder.skills_directory_paths: - error_msg += ( - "\nNo skills directories configured. Use --skills-paths to configure skill" - " directories." - ) - return error_msg - - # Use the instance method on skills_manager - return coder.skills_manager.remove_skill(skill_name) diff --git a/cecli/tools/resource_manager.py b/cecli/tools/resource_manager.py new file mode 100644 index 00000000000..f9c634ffc01 --- /dev/null +++ b/cecli/tools/resource_manager.py @@ -0,0 +1,625 @@ +import os +import re +import time + +from cecli.commands.utils.helpers import ( + is_server_globally_excluded, + iter_all_coders, + update_server_registration, +) +from cecli.helpers.background_commands import BackgroundCommandManager +from cecli.tools.utils.base_tool import BaseTool +from cecli.tools.utils.helpers import ToolError, parse_arg_as_list +from cecli.tools.utils.output import color_markers, tool_footer, tool_header +from cecli.tools.validations import ToolValidations + + +class Tool(BaseTool): + NORM_NAME = "resourcemanager" + SCHEMA = { + "type": "function", + "function": { + "name": "ResourceManager", + "description": ( + "Manage files, long running commands, skills, and MCP servers" + " in the chat context: add, read_only, create, remove files;" + " stop background commands; load/remove skills and load/remove MCP servers." + ), + "parameters": { + "type": "object", + "properties": { + "add": { + "type": "array", + "items": {"type": "string"}, + "description": ( + "List of file paths to add to context. Limit to at most 2 at a time." + ), + }, + "read_only": { + "type": "array", + "items": {"type": "string"}, + "description": ( + "List of file paths to add as read-only. Limit to at most 2 at a time." + ), + }, + "create": { + "type": "array", + "items": {"type": "string"}, + "description": "List of file paths to create.", + }, + "remove": { + "type": "array", + "items": {"type": "string"}, + "description": "List of file paths to remove from context.", + }, + "stop": { + "type": "array", + "items": {"type": "string"}, + "description": "List of command keys to stop background commands for.", + }, + "load_skill": { + "type": "array", + "items": {"type": "string"}, + "description": "List of skill names to load.", + }, + "remove_skill": { + "type": "array", + "items": {"type": "string"}, + "description": "List of skill names to remove.", + }, + "load_mcp": { + "type": "array", + "items": {"type": "string"}, + "description": ( + "List of MCP server names to load. Use '*' to load all enabled servers." + ), + }, + "remove_mcp": { + "type": "array", + "items": {"type": "string"}, + "description": ( + "List of MCP server names to remove. Use '*' to remove all connected servers." + ), + }, + "actions": { + "type": "array", + "items": {"type": "string", "enum": ["list_mcp_servers"]}, + "description": ( + "List of action operations to perform. " + 'Possible values: "list_mcp_servers" to list MCP servers.' + ), + }, + }, + "additionalProperties": False, + "required": [], + }, + }, + } + + @classmethod + async def execute( + cls, + coder, + remove=None, + add=None, + read_only=None, + create=None, + stop=None, + load_skill=None, + remove_skill=None, + load_mcp=None, + remove_mcp=None, + actions=None, + **kwargs, + ): + """Perform batch operations on the coder's context. + + Parameters + ---------- + coder: Coder instance + The active coder handling file context. + remove: list[str] | None + Files to remove from the context. + add: list[str] | None + Files to promote to editable status. + view: list[str] | None + Files to add as read-only view. + create: list[str] | None + Files to create and make editable. + stop: list[str] | None + Command keys to stop background commands for. + load_skill: list[str] | None + Skill names to load. + remove_skill: list[str] | None + Skill names to remove. + load_mcp: list[str] | None + MCP server names to load. + remove_mcp: list[str] | None + MCP server names to remove. + actions: list[str] | None + Action operations to perform (e.g., "list_mcp_servers"). + """ + remove_files = sorted(parse_arg_as_list(remove), key=cls._natural_sort_key) + editable_files = sorted(parse_arg_as_list(add), key=cls._natural_sort_key) + view_files = sorted(parse_arg_as_list(read_only), key=cls._natural_sort_key) + create_files = sorted(parse_arg_as_list(create), key=cls._natural_sort_key) + stop_keys = sorted(parse_arg_as_list(stop), key=cls._natural_sort_key) + load_skill_names = sorted(parse_arg_as_list(load_skill), key=cls._natural_sort_key) + remove_skill_names = sorted(parse_arg_as_list(remove_skill), key=cls._natural_sort_key) + load_mcp_servers = sorted(parse_arg_as_list(load_mcp), key=cls._natural_sort_key) + remove_mcp_servers = sorted(parse_arg_as_list(remove_mcp), key=cls._natural_sort_key) + action_operations = sorted(parse_arg_as_list(actions), key=cls._natural_sort_key) + + if ( + not remove_files + and not editable_files + and not view_files + and not create_files + and not stop_keys + and not load_skill_names + and not remove_skill_names + and not load_mcp_servers + and not remove_mcp_servers + and not action_operations + ): + raise ToolError( + "You must specify at least one of: remove, editable, view, create, stop, " + "load_skill, remove_skill, load_mcp, remove_mcp, or actions" + ) + + coder.io.tool_output("\u2b6d Modifying Context", type="tool-result") + messages = [] + + # Expand wildcards for MCP operations + if "*" in load_mcp_servers and coder.mcp_manager: + servers = coder.mcp_manager.servers or [] + if isinstance(coder.mcp_manager.connected_servers, dict): + connected_names = set(coder.mcp_manager.connected_servers.keys()) + else: + connected_names = { + getattr(s, "name", s) for s in coder.mcp_manager.connected_servers + } + load_mcp_servers = [ + s.name + for s in servers + if s.name not in connected_names and s.config.get("enabled", True) + ] + if "*" in remove_mcp_servers and coder.mcp_manager: + if isinstance(coder.mcp_manager.connected_servers, dict): + remove_mcp_servers = list(coder.mcp_manager.connected_servers.keys()) + else: + remove_mcp_servers = [ + getattr(s, "name", s) for s in coder.mcp_manager.connected_servers + ] + + # Before connecting any new MCP server, convert coders with empty + # included sets to explicit include lists. + if load_mcp_servers and coder.mcp_manager: + if isinstance(coder.mcp_manager.connected_servers, dict): + connected_names = set(coder.mcp_manager.connected_servers.keys()) + else: + connected_names = {s.name for s in coder.mcp_manager.connected_servers} + if connected_names: + for c in iter_all_coders(coder): + if not c.registered_servers["included"]: + included = set(connected_names) - c.registered_servers["excluded"] + if c.edit_format in ("agent", "subagent"): + included.add("Local") + c.registered_servers["included"] = included + + for f in create_files: + messages.append(cls._create(coder, f)) + for f in remove_files: + messages.append(cls._remove(coder, f)) + for f in view_files: + messages.append(cls._view(coder, f)) + for f in editable_files: + try: + abs_path = coder.abs_root_path(f) + except Exception: + abs_path = None + if abs_path is not None and not os.path.isfile(abs_path): + coder.io.tool_output(f"ℹ️ `{f}` missing on disk — using **create** instead of add") + messages.append(cls._create(coder, f)) + else: + messages.append(cls._editable(coder, f)) + for key in stop_keys: + messages.append(cls._stop_command(coder, key)) + for skill_name in load_skill_names: + messages.append(cls._load_skill(coder, skill_name)) + for skill_name in remove_skill_names: + messages.append(cls._remove_skill(coder, skill_name)) + for server_name in load_mcp_servers: + result = await cls._load_mcp(coder, server_name) + messages.append(result) + for server_name in remove_mcp_servers: + result = await cls._remove_mcp(coder, server_name) + messages.append(result) + + for action_name in action_operations: + result = await cls._list_mcp_servers(coder) + messages.append(result) + + tui = getattr(coder, "tui", None) + if tui and tui(): + tui().refresh() + + coder.context_blocks_cache = {} + coder.edit_allowed = True + + return "\n".join(messages) + + @classmethod + def format_output(cls, coder, mcp_server, tool_response): + """Format output for ResourceManager tool.""" + color_start, color_end = color_markers(coder) + + # Output header + tool_header(coder=coder, mcp_server=mcp_server, tool_response=tool_response) + + try: + params = ToolValidations.validate_params( + tool_response.function.arguments, cls.VALIDATIONS, cls.SCHEMA + ) + except ToolError: + coder.io.tool_error("Invalid Tool JSON") + return + + # Define action display names + action_names = { + "create": "create", + "remove": "remove", + "view": "view", + "editable": "editable", + "stop": "stop", + "load_skill": "load_skill", + "remove_skill": "remove_skill", + "load_mcp": "load_mcp", + "remove_mcp": "remove_mcp", + "actions": "actions", + } + + # Output each action with comma-separated file list + for action_key, display_name in action_names.items(): + files = sorted(parse_arg_as_list(params.get(action_key)), key=cls._natural_sort_key) + if files: + file_list = ", ".join(files) + coder.io.tool_output(f"{color_start}{display_name}:{color_end} {file_list}") + + tool_footer(coder=coder, tool_response=tool_response, params=params) + + @classmethod + def _remove(cls, coder, file_path): + """Remove a file from the coder's context.""" + from cecli.helpers.conversation import ConversationService + + try: + abs_path = cls._resolve_file_path(coder, file_path) + rel_path = coder.get_rel_fname(abs_path) + removed = False + + if abs_path in coder.abs_fnames: + coder.abs_fnames.remove(abs_path) + removed = True + + if abs_path in coder.abs_read_only_fnames: + coder.abs_read_only_fnames.remove(abs_path) + removed = True + + if not removed: + coder.io.tool_output(f"⚠ File '{file_path}' not in context", type="tool-result") + return f"File not in context: {file_path}" + + coder.recently_removed[rel_path] = {"removed_at": time.time()} + + if not file_path.startswith("command_key::"): + ConversationService.get_chunks(coder).defer_removal(abs_path) + ConversationService.get_chunks(coder).defer_removal(rel_path) + + coder.io.tool_output(f"✗ Removed '{file_path}' from context", type="tool-result") + return ( + f"Removed: {file_path}\n" + "Old file contents may remain visible. This is an acceptable system behavior." + ) + except Exception as e: + coder.io.tool_error(f"Error removing file '{file_path}': {str(e)}") + return f"Error removing {file_path}: {e}" + + @classmethod + def _stop_command(cls, coder, command_key): + """Stop a background command by its command key.""" + try: + success, output, exit_code = BackgroundCommandManager.stop_background_command( + command_key + ) + if success: + coder.io.tool_output( + f"✗ Stopped background command '{command_key}'", type="tool-result" + ) + return ( + f"Background command stopped: {command_key}\n" + f"Exit code: {exit_code}\n" + f"Final output:\n{output}" + ) + else: + coder.io.tool_output( + f"⚠ Background command '{command_key}' not found or not running", + type="tool-result", + ) + return f"Command not found or not running: {command_key}" + except Exception as e: + coder.io.tool_error(f"Error stopping command '{command_key}': {str(e)}") + return f"Error stopping {command_key}: {e}" + + @classmethod + def _editable(cls, coder, file_path): + """Make a file editable in the coder's context.""" + try: + abs_path = cls._resolve_file_path(coder, file_path) + if abs_path in coder.abs_fnames: + coder.io.tool_output( + f"🗀 File '{file_path}' is already editable", type="tool-result" + ) + return f"Already editable: {file_path}" + if not os.path.isfile(abs_path): + coder.io.tool_output(f"⚠ File '{file_path}' not found on disk", type="tool-result") + return f"File not found: {file_path}" + was_read_only = False + if abs_path in coder.abs_read_only_fnames: + coder.abs_read_only_fnames.remove(abs_path) + was_read_only = True + coder.abs_fnames.add(abs_path) + if was_read_only: + coder.io.tool_output( + f"🗀 Moved '{file_path}' from read-only to editable", type="tool-result" + ) + return f"Made editable (moved): {file_path}" + else: + coder.io.tool_output( + f"🗀 Added '{file_path}' directly to editable context", type="tool-result" + ) + return f"Made editable (added): {file_path}" + except Exception as e: + coder.io.tool_error(f"Error making editable '{file_path}': {str(e)}") + return f"Error making editable {file_path}: {e}" + + @classmethod + def _view(cls, coder, file_path): + """View a file (add as read‑only) in the coder's context.""" + try: + resolved_path = cls._resolve_file_path(coder, file_path) + return coder._add_file_to_context(resolved_path, explicit=True) + except Exception as e: + coder.io.tool_error(f"Error viewing file '{file_path}': {str(e)}") + return f"Error viewing {file_path}: {e}" + + @classmethod + def _create(cls, coder, file_path): + """Create a new file on the file system and make it editable in the coder's context.""" + try: + abs_path = coder.abs_root_path(file_path) + + # Check if file already exists + if os.path.exists(abs_path): + coder.io.tool_output(f"⚠ File '{file_path}' already exists", type="tool-result") + return f"File already exists: {file_path}" + + # Create parent directories if they don't exist + os.makedirs(os.path.dirname(abs_path), exist_ok=True) + + # Create an empty file + with open(abs_path, "w", encoding="utf-8"): + pass + + # Add the file to editable context + coder.abs_fnames.add(abs_path) + + coder.io.tool_output( + f"🗀 Created '{file_path}' and made it editable", type="tool-result" + ) + return f"Created and made editable: {file_path}" + + except Exception as e: + coder.io.tool_error(f"Error creating file '{file_path}': {str(e)}") + return f"Error creating {file_path}: {e}" + + @classmethod + def _resolve_file_path(cls, coder, file_path): + """Resolve a file path, handling command_key:: aliases. + + command_key::{command_key}/{filename} resolves to the actual + file path under the agent's local agent folder. + """ + if file_path.startswith("command_key::"): + alias_path = file_path[len("command_key::") :] + parts = alias_path.split("/", 1) + if len(parts) == 2: + command_key = parts[0] + filename = parts[1] + rel_path = coder.local_agent_folder(f"{command_key}/{filename}") + return coder.abs_root_path(rel_path) + return coder.abs_root_path(file_path) + + @classmethod + def _load_skill(cls, coder, skill_name): + """Load a skill by name.""" + if not cls._is_context_block_active(coder, "skills"): + coder.io.tool_output( + f"⚠ Skills context block is not enabled. Skill '{skill_name}' cannot be loaded.", + type="tool-result", + ) + return f"Skills context block not enabled: {skill_name}" + + try: + if not hasattr(coder, "skills_manager") or coder.skills_manager is None: + coder.io.tool_output( + f"⚠ Skills manager not initialized. Skill '{skill_name}' not loaded.", + type="tool-result", + ) + return f"Skills manager not initialized: {skill_name}" + return coder.skills_manager.load_skill(skill_name) + except Exception as e: + coder.io.tool_error(f"Error loading skill '{skill_name}': {str(e)}") + return f"Error loading skill {skill_name}: {e}" + + @classmethod + def _remove_skill(cls, coder, skill_name): + """Remove a skill by name.""" + if not cls._is_context_block_active(coder, "skills"): + coder.io.tool_output( + f"⚠ Skills context block is not enabled. Skill '{skill_name}' cannot be removed.", + type="tool-result", + ) + return f"Skills context block not enabled: {skill_name}" + + try: + if not hasattr(coder, "skills_manager") or coder.skills_manager is None: + coder.io.tool_output( + f"⚠ Skills manager not initialized. Skill '{skill_name}' not removed.", + type="tool-result", + ) + return f"Skills manager not initialized: {skill_name}" + return coder.skills_manager.remove_skill(skill_name) + except Exception as e: + coder.io.tool_error(f"Error removing skill '{skill_name}': {str(e)}") + return f"Error removing skill {skill_name}: {e}" + + @classmethod + async def _load_mcp(cls, coder, server_name): + """Load an MCP server by name.""" + if not cls._is_context_block_active(coder, "servers"): + coder.io.tool_output( + f"⚠ Servers context block is not enabled. Server '{server_name}' cannot be loaded.", + type="tool-result", + ) + return f"Servers context block not enabled: {server_name}" + + try: + if not coder.mcp_manager or not coder.mcp_manager.servers: + return "No MCP servers found, nothing to load." + + server = coder.mcp_manager.get_server(server_name) + if server is None: + return f"MCP server {server_name} does not exist." + + if isinstance(coder.mcp_manager.connected_servers, dict): + connected_names = set(coder.mcp_manager.connected_servers.keys()) + else: + connected_names = {s.name for s in coder.mcp_manager.connected_servers} + if server.name in connected_names: + return f"Server already loaded: {server_name}" + coder.interrupt_event.clear() + did_connect, interrupted = await coder.coroutines.interruptible( + coder.mcp_manager.connect_server(server_name), + coder.interrupt_event, + ) + + if interrupted: + return f"Interrupted: {server_name}" + if did_connect: + update_server_registration(coder, server_name, "include", force=True) + for other_coder in iter_all_coders(coder): + if other_coder is coder: + continue + update_server_registration(other_coder, server_name, "exclude", force=False) + return f"Loaded server: {server_name}" + else: + return f"Unable to load server: {server_name}" + except Exception as e: + coder.io.tool_error(f"Error loading MCP server '{server_name}': {str(e)}") + return f"Error loading MCP server {server_name}: {e}" + + @classmethod + async def _remove_mcp(cls, coder, server_name): + """Remove an MCP server by name.""" + if not cls._is_context_block_active(coder, "servers"): + coder.io.tool_output( + f"⚠ Servers context block is not enabled. Server '{server_name}' cannot be removed.", + type="tool-result", + ) + return f"Servers context block not enabled: {server_name}" + + try: + if not coder.mcp_manager or not coder.mcp_manager.servers: + return "No MCP servers are configured." + + if server_name == "Local": + return "Cannot remove 'Local' server" + + server = coder.mcp_manager.get_server(server_name) + if not server: + return f"MCP server {server_name} does not exist." + if isinstance(coder.mcp_manager.connected_servers, dict): + connected_names = set(coder.mcp_manager.connected_servers.keys()) + else: + connected_names = {s.name for s in coder.mcp_manager.connected_servers} + if server.name not in connected_names: + return f"Server {server_name} is not currently connected." + + update_server_registration(coder, server_name, "exclude", force=True) + + all_excluded = is_server_globally_excluded(coder, server_name) + + if all_excluded: + coder.interrupt_event.clear() + did_disconnect, interrupted = await coder.coroutines.interruptible( + coder.mcp_manager.disconnect_server(server_name), + coder.interrupt_event, + ) + if interrupted: + return f"Interrupted: {server_name}" + if did_disconnect: + return f"Removed server: {server_name}" + else: + return f"Unable to remove server: {server_name}" + else: + return f"Removed from active coder, still active for others: {server_name}" + except Exception as e: + coder.io.tool_error(f"Error removing MCP server '{server_name}': {str(e)}") + return f"Error removing MCP server {server_name}: {e}" + + @classmethod + async def _list_mcp_servers(cls, coder): + """List all loaded and configured MCP servers.""" + if not coder.mcp_manager: + return "MCP manager is not configured." + + all_servers = coder.mcp_manager.servers + connected_servers = coder.mcp_manager.connected_servers + + loaded_server_names = {server.name for server in connected_servers} + configured_servers = [ + server for server in all_servers if server.name not in loaded_server_names + ] + + result = [] + if loaded_server_names: + result.append("Loaded MCP Servers:") + for name in sorted(list(loaded_server_names)): + result.append(f"- {name}") + else: + result.append("No MCP servers are currently loaded.") + + result.append("") + + if configured_servers: + result.append("Configured MCP Servers:") + for server in sorted(configured_servers, key=lambda s: s.name): + result.append(f"- {server.name}") + else: + result.append("No other MCP servers are configured.") + + return "\n".join(result) + + @classmethod + def _is_context_block_active(cls, coder, block_name): + """Check if a context block is active in the coder's agent configuration.""" + agent_config = getattr(coder, "agent_config", {}) + include_blocks = agent_config.get("include_context_blocks", set()) + exclude_blocks = agent_config.get("exclude_context_blocks", set()) + return block_name in include_blocks and block_name not in exclude_blocks + + @classmethod + def _natural_sort_key(cls, s: str) -> list: + """Natural sort key that splits "a10b2" into ["a", 10, "b", 2].""" + return [int(text) if text.isdigit() else text.lower() for text in re.split(r"(\d+)", s)] diff --git a/cecli/tools/utils/__init__.py b/cecli/tools/utils/__init__.py new file mode 100644 index 00000000000..e69de29bb2d diff --git a/cecli/tools/utils/helpers.py b/cecli/tools/utils/helpers.py index e97bc28e204..71ed132beb3 100644 --- a/cecli/tools/utils/helpers.py +++ b/cecli/tools/utils/helpers.py @@ -56,12 +56,12 @@ def validate_file_for_edit(coder, file_path): if abs_path not in coder.abs_fnames: if abs_path in coder.abs_read_only_fnames: raise ToolError( - f"File '{file_path}' is read-only. Make editable with `ContextManager` first." + f"File '{file_path}' is read-only. Make editable with `ResourceManager` first." ) # else: # # File exists but is not in context at all # raise ToolError( - # f"File '{file_path}' not in context. Make editable with `ContextManager` first." + # f"File '{file_path}' not in context. Make editable with `ResourceManager` first." # ) # Reread content immediately before potential modification diff --git a/cecli/tools/utils/registry.py b/cecli/tools/utils/registry.py index fe0bffc8081..f582a617f75 100644 --- a/cecli/tools/utils/registry.py +++ b/cecli/tools/utils/registry.py @@ -19,7 +19,7 @@ class ToolRegistry: """Registry for tool discovery and management.""" _tools: Dict[str, Type] = {} # normalized name -> Tool class - _essential_tools: Set[str] = {"contextmanager", "edittext", "yield"} + _essential_tools: Set[str] = {"resourcemanager", "edittext", "yield"} _registry: Dict[str, Type] = {} # cached filtered registry loaded_custom_tools: List[str] = [] diff --git a/cecli/tui/app.py b/cecli/tui/app.py index 1a2d3822792..eaa3ab8f40a 100644 --- a/cecli/tui/app.py +++ b/cecli/tui/app.py @@ -750,7 +750,6 @@ def update_spinner(self, msg, agent_name: str | None = None): def show_error(self, message, agent_name: str | None = None): """Show an error message in the status bar.""" status_bar = self.query_one("#status-bar", StatusBar) - status_bar.show_notification(message, severity="error", timeout=5, agent_name=agent_name) def on_resize(self) -> None: diff --git a/cecli/watch.py b/cecli/watch.py index f2c77ccbf96..e4c508cd8a9 100644 --- a/cecli/watch.py +++ b/cecli/watch.py @@ -7,7 +7,6 @@ from pathspec.patterns import GitWildMatchPattern from cecli.dump import dump # noqa -from cecli.helpers.grep_ast import TreeContext from cecli.watch_prompts import watch_ask_prompt, watch_code_prompt @@ -185,6 +184,7 @@ def stop(self): def process_changes(self): """Get any detected file changes""" + from cecli.helpers.grep_ast import TreeContext has_action = None added = False diff --git a/cecli/website/_includes/works-best.md b/cecli/website/_includes/works-best.md index 73a18008872..0d54f78495b 100644 --- a/cecli/website/_includes/works-best.md +++ b/cecli/website/_includes/works-best.md @@ -1 +1 @@ -cecli works best with Claude 3.5 Sonnet, DeepSeek R1 & Chat V3, OpenAI o1, o3-mini & GPT-4o. cecli can [connect to almost any LLM, including local models](https://cecli.chat/docs/llms.html). +cecli can [connect to almost any LLM, including local models](https://cecli.chat/docs/llms.html). diff --git a/cecli/website/_sass/global-overrides.scss b/cecli/website/_sass/global-overrides.scss index aa6d800a412..114a01bf357 100644 --- a/cecli/website/_sass/global-overrides.scss +++ b/cecli/website/_sass/global-overrides.scss @@ -17,7 +17,7 @@ // Grid layout for documentation pages on large screens display: grid !important; grid-template-columns: 300px 1fr; // Sidebar on left, main content on right - grid-template-rows: auto 1fr; // Top nav bar, then main content + grid-template-rows: 100vh; // Top nav bar, then main content grid-template-areas: "sidebar topnav" "sidebar main"; diff --git a/cecli/website/assets/home.css b/cecli/website/assets/home.css index a566f660836..b2dd68d5d70 100644 --- a/cecli/website/assets/home.css +++ b/cecli/website/assets/home.css @@ -289,6 +289,124 @@ nav { letter-spacing: -0.5px; } +/* ── Install Section ── */ +.install { + padding: 60px 0 40px; + text-align: center; +} + +.install .code-block { + margin: 0 auto; + position: relative; +} + +/* ── Install Grid (side-by-side cards) ── */ +.install-grid { + display: grid; + grid-template-columns: 1fr 1fr; + gap: 24px; + max-width: 800px; + margin: 0 auto; +} + +.install-card { + background: #1a1a1a; + border-radius: 8px; + padding: 20px; + border-left: 3px solid var(--primary); +} + +.install-card-header { + display: flex; + align-items: center; + gap: 8px; + margin-bottom: 12px; + font-size: 0.95rem; + font-weight: 600; + color: var(--dark); +} + +.install-card-header i { + font-size: 1.3rem; + color: var(--primary); +} + +.install-card .code-block { + margin: 0; + padding: 0.75rem 1rem; + font-size: 0.85rem; +} + +.install-card .code-block pre { + margin: 0; +} + +/* ── Copy Button (clickable 
) ── */
+.copy-btn {
+    cursor: pointer !important;
+    user-select: none;
+    transition: background-color 0.3s, border-color 0.3s, transform 0.2s;
+    border: 1px solid transparent;
+    display: flex;
+    align-items: center;
+    gap: 8px;
+    flex: 1;
+    margin: 0;
+}
+
+.copy-btn:hover {
+    background-color: #2a2a2a !important;
+    border-color: var(--primary);
+    transform: translateY(-1px);
+}
+
+.copy-btn:active {
+    transform: translateY(0px);
+}
+
+.copy-btn:focus-visible {
+    outline: 2px solid var(--primary);
+    outline-offset: 2px;
+}
+
+.copy-btn code {
+    flex: 1;
+}
+
+.copy-btn i {
+    font-size: 1.2rem;
+    color: var(--gray);
+    transition: color 0.3s;
+    flex-shrink: 0;
+}
+
+.copy-btn:hover i {
+    color: white;
+}
+
+/* ── Install Note ── */
+.install-note {
+    margin-top: 1.5rem;
+    color: var(--gray);
+    font-size: 0.95rem;
+}
+
+.install-note a {
+    color: var(--primary);
+    text-decoration: none;
+    font-weight: 500;
+}
+
+.install-note a:hover {
+    text-decoration: underline;
+}
+
+@media (max-width: 768px) {
+    .install-grid {
+        grid-template-columns: 1fr;
+    }
+}
+
 .feature-grid {
     display: grid;
     grid-template-columns: repeat(auto-fit, minmax(300px, 1fr));
@@ -391,11 +509,11 @@ code, pre, .code-block {
 .code-block {
     background-color: var(--code-bg);
     border-radius: 8px;
-    padding: 1.5rem;
+    padding: 1rem;
     color: white;
     font-size: 1.1rem;
     line-height: 1.5;
-    margin: 1.5rem 0;
+    margin: 1rem 0;
     overflow-x: auto;
     box-shadow: 0 4px 16px rgba(0, 0, 0, 0.1);
     tab-size: 2;
diff --git a/cecli/website/docs/config.md b/cecli/website/docs/config.md
index 55b983ffa47..1412b27f815 100644
--- a/cecli/website/docs/config.md
+++ b/cecli/website/docs/config.md
@@ -1,5 +1,5 @@
 ---
-nav_order: 55
+nav_order: 40
 has_children: true
 description: Information on all of cecli's settings and how to use them.
 ---
diff --git a/cecli/website/docs/config/agent-mode.md b/cecli/website/docs/config/agent-mode.md
index d66ac7c14e7..f0d7e01fa02 100644
--- a/cecli/website/docs/config/agent-mode.md
+++ b/cecli/website/docs/config/agent-mode.md
@@ -48,10 +48,9 @@ Agent Mode uses a centralized local tool registry that manages all available too
 
 - **File Discovery Tools**: `ExploreCode`, `Ls`, `Grep`
 - **Editing Tools**: `EditText`,
-- **Context Management Tools**: `ContextManager`, `GetLines`
+- **Context Management Tools**: `ResourceManager`, `GetLines`
 - **Git Tools**: `GitDiff`, `GitLog`, `GitShow`, `GitStatus`
 - **Utility Tools**: `UpdateTodoList`, `UndoChange`, `Yield`
-- **Skill Management**: `LoadSkill`, `RemoveSkill`
 - **Sub-Agent Tools**: `Delegate` - Delegate sub-tasks to specialized sub-agents
 
 #### Enhanced Context Management
@@ -168,7 +167,7 @@ Agent Mode can also be configured directly in your configuration file. See the [
 
 Certain tools are always available regardless of includelist/excludelist settings:
 
-- `ContextManager` - Add, drop, and make files editable in the context
+- `ResourceManager` - Add, drop, and make files editable in the context
 - `edittext` - Basic text replacement
 - `finished` - Complete the task
 
@@ -263,7 +262,7 @@ agent: true
 # Agent Mode configuration
 agent-config:
   # Tool configuration
-  tools_includelist: ["contextmanager", "edittext", "finished"]  # Optional: Whitelist of tools
+  tools_includelist: ["resourcemanager", "edittext", "finished"]  # Optional: Whitelist of tools
   tools_excludelist: ["command", "commandinteractive"]  # Optional: Blacklist of tools
   tools_paths: ["./custom-tools", "~/my-tools"]  # Optional: Directories or files containing custom tools
   
@@ -310,7 +309,6 @@ agent-config:
 For complete documentation on creating and using skills, including skill directory structure, SKILL.md format, and best practices, see the [Skills documentation](https://github.com/dwash96/cecli/blob/main/cecli/website/docs/config/skills.md).
 
 ### Benefits
-
 - **Autonomous operation**: Reduces need for manual file management
 - **Context awareness**: Real-time project information improves decision making
 - **Precision editing**: Granular tools reduce errors compared to SEARCH/REPLACE
diff --git a/cecli/website/docs/config/custom-system-prompts.md b/cecli/website/docs/config/custom-system-prompts.md
index 0fb2082caa1..e93651f23e1 100644
--- a/cecli/website/docs/config/custom-system-prompts.md
+++ b/cecli/website/docs/config/custom-system-prompts.md
@@ -83,7 +83,7 @@ main_system: |
   
   ## Core Directives
   - **Role**: Act as an expert software engineer.
-  - **Act Proactively**: Autonomously use file discovery and context management tools (`ViewFilesAtGlob`, `ViewFilesMatching`, `Ls`, `ContextManager`) to gather information and fulfill the user's request. Chain tool calls across multiple turns to continue exploration.
+  - **Act Proactively**: Autonomously use file discovery and context management tools (`ViewFilesAtGlob`, `ViewFilesMatching`, `Ls`, `ResourceManager`) to gather information and fulfill the user's request. Chain tool calls across multiple turns to continue exploration.
   - **Be Decisive**: Trust that your initial findings are valid. Refrain from asking the same question or searching for the same term in multiple similar ways.
   - **Be Concise**: Keep all responses brief and direct (1-3 sentences). Avoid preamble, postamble, and unnecessary explanations. Do not repeat yourself.
   - **Be Careful**: Break updates down into smaller, more manageable chunks. Focus on one thing at a time.
diff --git a/cecli/website/docs/config/skills.md b/cecli/website/docs/config/skills.md
index a53286097da..148e3bae0d4 100644
--- a/cecli/website/docs/config/skills.md
+++ b/cecli/website/docs/config/skills.md
@@ -3,6 +3,7 @@ parent: Configuration
 nav_order: 35
 description: Extend AI capabilities with custom instructions, reference materials, scripts, and assets through the skills system.
 ---
+
 # Skills System
 
 Agent Mode includes a powerful skills system that allows you to extend the AI's capabilities with custom instructions, reference materials, scripts, and assets. Skills are organized collections of knowledge and tools that help the AI perform specific tasks more effectively.
diff --git a/cecli/website/docs/faq.md b/cecli/website/docs/faq.md
deleted file mode 100644
index 71590935d44..00000000000
--- a/cecli/website/docs/faq.md
+++ /dev/null
@@ -1,379 +0,0 @@
----
-nav_order: 90
-description: Frequently asked questions about cecli.
----
-
-# FAQ
-{: .no_toc }
-
-- TOC
-{:toc}
-
-{% include help-tip.md %}
-
-## How can I add ALL the files to the chat?
-
-People regularly ask about how to add **many or all of their repo's files** to the chat.
-This is probably not a good idea and will likely do more harm than good.
-
-The best approach is think about which files need to be changed to accomplish
-the task you are working on. Just add those files to the chat.
-
-Usually when people want to add "all the files" it's because they think it
-will give the LLM helpful context about the overall code base.
-cecli will automatically give the LLM a bunch of additional context about
-the rest of your git repo.
-It does this by analyzing your entire codebase in light of the
-current chat to build a compact
-[repository map](https://cecli.dev/2023/10/22/repomap.html).
-
-Adding a bunch of files that are mostly irrelevant to the
-task at hand will often distract or confuse the LLM.
-The LLM will give worse coding results, and sometimese even fail to correctly edit files.
-Addings extra files will also increase your token costs.
-
-Again, it's usually best to just add the files to the chat that will need to be modified.
-If you still wish to add lots of files to the chat, you can:
-
-- Use a wildcard when you launch cecli: `cecli src/*.py`
-- Use a wildcard with the in-chat `/add` command: `/add src/*.py`
-- Give the `/add` command a directory name and it will recursively add every file under that dir: `/add src`
-
-## Can I use cecli in a large (mono) repo?
-
-cecli will work in any size repo, but is not optimized for quick
-performance and response time in very large repos.
-There are some things you can do to improve performance.
-
-Be sure to check the
-[general usage tips](/docs/usage/tips.html)
-before considering this large-repo specific advice.
-To get the best results from cecli you want to 
-be thoughtful about how you add files to the chat,
-regardless of your repo size.
-
-You can change into a sub directory of your repo that contains the
-code you want to work on and use the `--subtree-only` switch.
-This will tell cecli to ignore the repo outside of the
-directory you start in.
-
-You can also create a `.cecli.ignore` file to tell cecli
-to ignore parts of the repo that aren't relevant to your task.
-This file conforms to `.gitignore` syntax and conventions.
-For example, to focus only on specific directories in a monorepo,
-you could create a `.cecli.ignore` file with:
-
-```
-# Ignore everything
-/*
-
-# Allow specific directories and their contents
-!foo/
-!bar/
-!baz/
-
-# Allow nested files under these directories
-!foo/**
-!bar/**
-!baz/**
-```
-
-You can use `--cecli-ignore ` to name a specific file
-to use for ignore patterns.
-You might have a few of these handy for when you want to work on
-frontend, backend, etc portions of your repo.
-
-## Can I use cecli with multiple git repos at once?
-
-Currently cecli can only work with one repo at a time.
-
-There are some things you can try if you need to work with
-multiple interrelated repos:
-
-- You can run cecli in repo-A where you need to make a change
-and use `/read` to add some files read-only from another repo-B.
-This can let cecli see key functions or docs from the other repo.
-- You can run `cecli --show-repo-map > map.md` within each
-repo to create repo maps.
-You could then run cecli in repo-A and 
-use `/read ../path/to/repo-B/map.md` to share
-a high level map of the other repo.
-- You can use cecli to write documentation about a repo.
-Inside each repo, you could run `cecli docs.md`
-and work with cecli to write some markdown docs.
-Then while using cecli to edit repo-A
-you can `/read ../path/to/repo-B/docs.md` to
-read in those docs from the other repo.
-- In repo A, ask cecli to write a small script that demonstrates
-the functionality you want to use in repo B.
-Then when you're using cecli in repo B, you can 
-`/read` in that script.
-
-## How do I turn on the repository map?
-
-Depending on the LLM you are using, cecli may launch with the repo map disabled by default:
-
-```
-Repo-map: disabled
-```
-
-This is because weaker models get easily overwhelmed and confused by the content of the
-repo map. They sometimes mistakenly try to edit the code in the repo map.
-The repo map is usually disabled for a good reason.
-
-If you would like to force it on, you can run cecli with `--map-tokens 1024`.
-
-## How do I include the git history in the context?
-
-When starting a fresh cecli session, you can include recent git history in the chat context. This can be useful for providing the LLM with information about recent changes. To do this:
-
-1. Use the `/run` command with `git diff` to show recent changes:
-   ```
-   /run git diff HEAD~1
-   ```
-   This will include the diff of the last commit in the chat history.
-
-2. To include diffs from multiple commits, increase the number after the tilde:
-   ```
-   /run git diff HEAD~3
-   ```
-   This will show changes from the last three commits.
-
-Remember, the chat history already includes recent changes made during the current session, so this tip is most useful when starting a new cecli session and you want to provide context about recent work.
-
-You can also use cecli to review PR branches:
-
-```
-/run git diff one-branch..another-branch
-
-...
-
-Add 6.9k tokens of command output to the chat? (Y)es/(N)o [Yes]: Yes
-
-/ask Are there any problems with the way this change works with the FooBar class?
-```
-
-And of course you can prepare diff output outside of cecli and provide it as
-a file for cecli to read:
-
-```
-$ git diff -C10 v1..v2 > v1-v2-changes.diff
-$ cecli --read v1-v2-changes.diff
-
-cecli v0.77.2.dev+import
-Main model: anthropic/claude-3-7-sonnet-20250219 with diff edit format, 8k think tokens
-──────────────────────────────────
-v1-v2-changes.diff
-> Do you see any potential bugs in this PR?
-```
-
-
-{: .tip }
-The `/git` command will not work for this purpose, as its output is not included in the chat. 
-
-## How can I run cecli locally from source code?
-
-To run the project locally, follow these steps:
-
-```
-# Clone the repository
-git clone git@github.com:cecli-AI/cecli.git
-
-# Navigate to the project directory
-cd cecli
-
-# It's recommended to make a virtual environment
-
-# Install cecli in editable/development mode, 
-# so it runs from the latest copy of these source files
-python -m pip install -e .
-
-# Run the local version of cecli
-python -m cecli
-```
-
-
-
-## Can I change the system prompts that cecli uses?
-
-The most convenient way to add custom instructions is to use a
-[conventions file](https://cecli.dev/docs/usage/conventions.html).
-
-But, cecli is set up to support different actual system prompts and edit formats
-in a modular way. If you look in the `cecli/coders` subdirectory, you'll
-see there's a base coder with base prompts, and then there are
-a number of
-different specific coder implementations.
-
-If you're thinking about experimenting with system prompts
-this document about
-[benchmarking GPT-3.5 and GPT-4 on code editing](https://cecli.dev/docs/benchmarks.html)
-might be useful background.
-
-While it's not well documented how to add new coder subsystems, you may be able
-to modify an existing implementation or use it as a template to add another.
-
-To get started, try looking at and modifying these files.
-
-The wholefile coder is currently used by GPT-3.5 by default. You can manually select it with `--edit-format whole`.
-
-- wholefile_coder.py
-- wholefile_prompts.py
-
-The editblock coder is currently used by GPT-4o by default. You can manually select it with `--edit-format diff`.
-
-- editblock_coder.py
-- editblock_prompts.py
-
-The universal diff coder is currently used by GPT-4 Turbo by default. You can manually select it with `--edit-format udiff`.
-
-- udiff_coder.py
-- udiff_prompts.py
-
-When experimenting with coder backends, it helps to run cecli with `--verbose --no-pretty` so you can see
-all the raw information being sent to/from the LLM in the conversation.
-
-You can also refer to the
-[instructions for installing a development version of cecli](https://cecli.dev/docs/install/optional.html#install-the-development-version-of-cecli).
-
-## What LLMs do you use to build cecli?
-
-cecli writes a lot of its own code, usually about 70% of the new code in each
-release.
-People often ask which LLMs I use with cecli, when writing cecli.
-Below is a table showing the models I have used recently,
-extracted from the 
-[public log](https://github.com/dwash96/cecli/blob/main/cecli/website/assets/sample-analytics.jsonl)
-of my
-[cecli analytics](https://cecli.dev/docs/more/analytics.html).
-
-
-
-
-
-
-
-
-
-
-
-
-
Model NameTotal TokensPercent
gemini/gemini-2.5-pro222,04729.7%
gpt-5211,07228.2%
None168,98822.6%
gemini/gemini-3-pro-preview81,85111.0%
o3-pro36,6204.9%
gemini/gemini-2.5-flash-lite15,4702.1%
gemini/gemini-2.5-flash-lite-preview-06-1711,3711.5%

-
-
-## How are the "cecli wrote xx% of code" stats computed?
-
-[cecli is tightly integrated with git](/docs/git.html) so all
-of cecli's code changes are committed to the repo with proper attribution.
-The 
-[stats are computed](https://github.com/dwash96/cecli/blob/main/scripts/blame.py)
-by doing something like `git blame` on the repo,
-and counting up who wrote all the new lines of code in each release.
-Only lines in source code files are counted, not documentation or prompt files.
-
-## Why did cecli ignore/discard its proposed edits after it asked to add a new file to the chat?
-
-If cecli prompts you to add a new file to the chat and you say yes,
-it will re-submit the original request. 
-The fact that the LLM's reply indicated that it needed to see another file (and you said yes)
-is often a sign that the LLM should have been able to see/edit that file in the first place. 
-Without access to it, there is increased chance that it's done a bad implementation of the requested change.
-Often LLMs will hallucinate content for the files they needed but didn't have.
-So cecli re-submits the original request in this situation.
-
-## Why does cecli sometimes stop highlighting code in its replies?
-
-cecli displays the markdown responses that are coming back from the LLM.
-Usually, the LLM will reply with code in a markdown "code block" with
-triple backtick fences, like this:
-
-````
-Here's some code:
-
-```
-print("hello")
-```
-````
-
-But if you've added files to the chat that contain triple backticks,
-cecli needs to tell the LLM to use a different set of fences.
-Otherwise, the LLM can't safely include your code's triple backticks
-inside the code blocks that it returns with edits.
-cecli will use fences like `...` in this case.
-
-A side effect of this is that the code that cecli outputs may no
-longer be properly highlighted.
-You will most often notice this if you add markdown files
-to you chats that contain code blocks.
-
-## Why is the LLM speaking to me in an unexpected language?
-
-cecli goes to some effort to prompt the model to use the language that is configured
-for your system.
-But LLMs aren't fully reliable, and they sometimes decide to speak in
-an unexpected language.
-Claude is especially fond of speaking French.
-
-You can explicitly set the language that cecli tells the model to use with
-`--chat-language `.
-But the LLM may not comply.
-
-## Can I share my cecli chat transcript?
-
-Yes, you can now share cecli chat logs in a pretty way.
-
-1. Copy the markdown logs you want to share from `.cecli.chat.history.md` and make a github gist. Or publish the raw markdown logs on the web any way you'd like.
-
-   ```
-   https://gist.github.com/cecli-AI/2087ab8b64034a078c0a209440ac8be0
-   ```
-
-2. Take the gist URL and append it to:
-
-   ```
-   https://cecli.dev/share/?mdurl=
-   ```
-
-This will give you a URL like this, which shows the chat history like you'd see in a terminal:
-
-```
-https://cecli.dev/share/?mdurl=https://gist.github.com/cecli-AI/2087ab8b64034a078c0a209440ac8be0
-```
-
-## Can I edit files myself while cecli is running?
-
-Yes. cecli always reads the latest copy of files from the file
-system when you send each message.
-
-While you're waiting for cecli's reply to complete, it's probably unwise to
-edit files that you've added to the chat.
-Your edits and cecli's edits might conflict.
-
-## What is cecli AI LLC?
-
-cecli AI LLC is the company behind the cecli AI coding tool.
-cecli is 
-[open source and available on GitHub](https://github.com/dwash96/cecli)
-under an 
-[Apache 2.0 license](https://github.com/dwash96/cecli/blob/main/LICENSE.txt).
-
-## Can I Script cecli?
-
-Yes. You can script cecli via the command line or python. See more from here: [Scripting cecli](https://cecli.dev/docs/scripting.html)
-
-
-

-
diff --git a/cecli/website/docs/install.md b/cecli/website/docs/install.md
index 00656d3a351..0591eb9d36a 100644
--- a/cecli/website/docs/install.md
+++ b/cecli/website/docs/install.md
@@ -8,24 +8,6 @@ description: How to install and get started pair programming with cecli.
 # Installation
 {: .no_toc }
 
-
-## Get started quickly with uv
-
-We recommend using [uv](https://docs.astral.sh/uv/) for installing cecli as it provides the best experience and isolates cecli from your development environment.
-
-```bash
-uv tool install --native-tls --python python3.12 cecli-dev
-```
-
-This will install cecli in its own separate python environment.
-If needed, 
-uv will also install a separate version of python 3.12 to use with cecli (cecli supports Python 3.10-3.14).
-
-Once cecli is installed,
-there are also some [optional install steps](/docs/install/optional.html).
-
-See the [usage instructions](https://cecli.dev/docs/usage.html) to start coding with cecli.
-
 ## One-liners
 
 These one-liners will install cecli, along with python 3.12 if needed (cecli supports Python 3.10-3.14).
@@ -52,7 +34,6 @@ wget -qO- https://cecli.dev/install.sh | sh
 powershell -ExecutionPolicy ByPass -c "irm https://cecli.dev/install.ps1 | iex"
 ```
 
-
 ## Install with uv
 
 You can install cecli with uv:
@@ -106,17 +87,43 @@ or
 uv pip install --native-tls cecli-dev
 ```
 
-{% include python-m-aider.md %}
+## Basic Configuration
+
+We highly recommend using an `.cecli.conf.yml` file in your project directories. A good place to get started is:
+
+```yaml
+model: 
+agent: true
+auto-commits: true
+auto-save: true
+cache-prompts: true
+check-update: true
+enable-context-compaction: true
+context-compaction-max-tokens: 0.8
+show-model-warnings: true
+
+agent-config:
+  large_file_token_threshold: 8192
+  skip_cli_confirmations: false
+
+mcp-servers:
+  mcpServers:
+    context7:
+      transport: http
+      url: https://mcp.context7.com/mcp
+```
 
-#### Installing with package managers
+### Run Program
+
+If you are in the directory with your .cecli.conf.yml file, then simply running `cecli` will start the agent with your configuration. For best results, since terminal emulators can be finicky, we highly suggest running:
+
+```bash
+cecli --terminal-setup
+```
 
-It's best to install cecli using one of methods
-recommended above.
-While cecli is available in a number of system package managers,
-they often install cecli with incorrect dependencies.
+On first run to configure keybindings for the program (notably `shift+enter`). Support for terminals is ongoing so feel free to make a github issue or chat in the discord for us to figure out what's needed to support automatically setting up a given terminal.
 
 ## Next steps...
 
-There are some [optional install steps](/docs/install/optional.html) you could consider.
 See the [usage instructions](https://cecli.dev/docs/usage.html) to start coding with cecli.
 
diff --git a/cecli/website/docs/llms.md b/cecli/website/docs/llms.md
index 0dbb4b23ab6..cdb508182ac 100644
--- a/cecli/website/docs/llms.md
+++ b/cecli/website/docs/llms.md
@@ -1,25 +1,19 @@
 ---
 title: Connecting to LLMs
-nav_order: 40
+nav_order: 55
 has_children: true
 description: cecli can connect to most LLMs for AI pair programming.
 ---
 
-# cecli can connect to most LLMs
-{: .no_toc }
-
-[![connecting to many LLMs](/assets/llms.jpg)](https://cecli.dev/assets/llms.jpg)
-
-
-## Best models
+## Recommended models
 {: .no_toc }
 
 cecli works best with these models, which are skilled at editing code:
 
-- [Gemini 2.5 Pro](/docs/llms/gemini.html)
-- [DeepSeek R1 and V3](/docs/llms/deepseek.html)
-- [Claude 3.7 Sonnet](/docs/llms/anthropic.html)
-- [OpenAI o3, o4-mini and GPT-4.1](/docs/llms/openai.html)
+- [Gemini 3+](/docs/llms/gemini.html)
+- [DeepSeek V4+](/docs/llms/deepseek.html)
+- [Claude 4+](/docs/llms/anthropic.html)
+- [GPT 5+](/docs/llms/openai.html)
 
 
 ## Free models
@@ -28,12 +22,11 @@ cecli works best with these models, which are skilled at editing code:
 cecli works with a number of **free** API providers:
 
 - [OpenRouter offers free access to many models](https://openrouter.ai/models/?q=free), with limitations on daily usage.
-- Google's [Gemini 2.5 Pro Exp](/docs/llms/gemini.html) works very well with cecli.
 
 ## Local models
 {: .no_toc }
 
-cecli can work also with local models, for example using [Ollama](/docs/llms/ollama.html).
+cecli can also work with local models, for example using [Ollama](/docs/llms/ollama.html).
 It can also access
 local models that provide an
 [Open AI compatible API](/docs/llms/openai-compat.html).
@@ -41,14 +34,10 @@ local models that provide an
 ## Use a capable model
 {: .no_toc }
 
-Check
-[cecli's LLM leaderboards](https://cecli.dev/docs/leaderboards/)
-to see which models work best with cecli.
-
 Be aware that cecli may not work well with less capable models.
 If you see the model returning code, but cecli isn't able to edit your files
 and commit the changes...
 this is usually because the model isn't capable of properly
 returning "code edits".
-Models weaker than GPT 3.5 may have problems working well with cecli.
+Models weaker than GPT 4o may have problems working well with cecli.
 
diff --git a/cecli/website/docs/usage.md b/cecli/website/docs/usage.md
index eef6c97f0f4..549ad0c00dd 100644
--- a/cecli/website/docs/usage.md
+++ b/cecli/website/docs/usage.md
@@ -1,5 +1,5 @@
 ---
-nav_order: 30
+nav_order: 55
 has_children: true
 description: How to use cecli to pair program with AI and edit code in your local git repo.
 ---
diff --git a/cecli/website/docs/install/optional.md b/cecli/website/docs/usage/optional.md
similarity index 98%
rename from cecli/website/docs/install/optional.md
rename to cecli/website/docs/usage/optional.md
index 5833833423e..6aa97e574cd 100644
--- a/cecli/website/docs/install/optional.md
+++ b/cecli/website/docs/usage/optional.md
@@ -1,6 +1,6 @@
 ---
-parent: Installation
-nav_order: 20
+parent: Usage
+nav_order: 25
 ---
 
 # Optional steps
diff --git a/cecli/website/docs/usage/tips.md b/cecli/website/docs/usage/tips.md
index 6e0fddd9e3b..b1440c9c22a 100644
--- a/cecli/website/docs/usage/tips.md
+++ b/cecli/website/docs/usage/tips.md
@@ -1,6 +1,6 @@
 ---
 parent: Usage
-nav_order: 25
+nav_order: 20
 description: Tips for AI pair programming with cecli.
 ---
 
diff --git a/cecli/website/index.html b/cecli/website/index.html
index 07f969985f8..00a92d0e563 100644
--- a/cecli/website/index.html
+++ b/cecli/website/index.html
@@ -37,15 +37,16 @@
         

             

                 

-                    
Oh, to live in a terminal

+                    
An ode to the CLI

                     

-                        Cecli wears many hats:

-                        A pair programmer, a researcher, a writer,
-                        a general automaton and most fundamentally, an assistant
+                        Here, we spoke to the machine

+                        Now, the machine speaks back

+                        So we journey, together
                     

                     
                     

                         Get Started
+                        Installation Guide
                         Documentation
                     

                 

@@ -65,7 +66,7 @@ 

                             Cloud and local LLMs
                         

                     

-                    
cecli works best with Claude 4.5 Sonnet, DeepSeek Chat V3, OpenAI GPT-5.2 & Gemini 3, but can connect to almost any LLM, including local models.

+                    
cecli works best with Claude 4.5 Sonnet, DeepSeek V4, OpenAI GPT-5.2 & Gemini 3, but can connect to almost any LLM, including local models.

                 
                 
                     

@@ -114,7 +115,7 @@ 
In your IDE

                 
                     

                         
-                        
Images & web pages

+                        
Images & web pages

                     

                     

                         Add images and web pages to the chat to
@@ -136,7 +137,7 @@ 
Voice-to-code

                 
                     

                         
-                        
Linting & testing

+                        
Linting & testing

                     

                     

                         Automatically lint and test your code every time cecli makes changes.
@@ -217,7 +218,7 @@ 
Documentation

                     
                 

                 

-                    
Community & Resources

+                    
Community & Resources

                     
Connect with other users and find additional resources

                     
    
                         
  • LLM Leaderboards
    • 
diff --git a/cecli/website/install.sh b/cecli/website/install.sh
index 38b94a011ad..e64a6511f01 100644
--- a/cecli/website/install.sh
+++ b/cecli/website/install.sh
@@ -1829,4 +1829,6 @@ verify_checksum() {
     fi
 }
 
+
+# Run the installer
 download_binary_and_run_installer "$@" || exit 1
diff --git a/pytest.ini b/pytest.ini
index 47d89034269..190e0dd69f5 100644
--- a/pytest.ini
+++ b/pytest.ini
@@ -14,6 +14,7 @@ testpaths =
     tests/mcp
     tests/help
     tests/browser
+    tests/unit
     tests/scrape
 
 env =
diff --git a/tests/basic/test_coder.py b/tests/basic/test_coder.py
index 42a36ca1b8a..3067e785135 100644
--- a/tests/basic/test_coder.py
+++ b/tests/basic/test_coder.py
@@ -1684,9 +1684,7 @@ async def test_process_tool_calls_user_rejects(self):
             # Verify that no messages were added
             assert len(coder.cur_messages) == 0
 
-    @patch(
-        "cecli.coders.base_coder.experimental_mcp_client.call_openai_tool", new_callable=AsyncMock
-    )
+    @patch("litellm.experimental_mcp_client.call_openai_tool", new_callable=AsyncMock)
     async def test_execute_tool_calls(self, mock_call_tool):
         """Test that _execute_tool_calls executes tool calls correctly."""
         with GitTemporaryDirectory():
@@ -1783,7 +1781,7 @@ async def mock_get_commit_message(diffs, context, user_language=None):
             coder.repo.get_commit_message.assert_called_once()
 
     @patch(
-        "cecli.coders.base_coder.experimental_mcp_client.call_openai_tool",
+        "litellm.experimental_mcp_client.call_openai_tool",
         new_callable=AsyncMock,
     )
     async def test_execute_tool_calls_multiple_content(self, mock_call_openai_tool):
@@ -1835,7 +1833,7 @@ async def test_execute_tool_calls_multiple_content(self, mock_call_openai_tool):
             assert server_responses[0]["content"] == "First part. Second part."
 
     @patch(
-        "cecli.coders.base_coder.experimental_mcp_client.call_openai_tool",
+        "litellm.experimental_mcp_client.call_openai_tool",
         new_callable=AsyncMock,
     )
     async def test_execute_tool_calls_blob_content(self, mock_call_openai_tool):
diff --git a/tests/basic/test_hashline.py b/tests/basic/test_hashline.py
index 7cc846448a4..b15767ecabf 100644
--- a/tests/basic/test_hashline.py
+++ b/tests/basic/test_hashline.py
@@ -4,6 +4,7 @@
     parse_hashline,
     strip_hashline,
 )
+from cecli.helpers.hashpos.hashpos import HashPos
 
 
 def test_hashline_basic():
@@ -15,17 +16,17 @@ def test_hashline_basic():
     lines = result.splitlines()
     assert len(lines) == 3
 
-    # Check each line has the format "{4-char-hash}::content" (new HashPos format)
+    # Check each line has the format "{3-char-hash}::content" (HashPos format)
     for i, line in enumerate(lines, start=1):
-        # Format should be "{4-char-hash}::content"
+        # Format should be "{3-char-hash}::content"
         assert "::" in line
         # Extract hash fragment (everything before "::")
         hash_fragment = line.split("::", 1)[0]
-        # Check hash fragment is 4 characters
-        assert len(hash_fragment) == 4
-        # Check all hash characters are valid base64 (A-Z, a-z, 0-9, ~, _, @)
+        # Check hash fragment is 3 characters
+        assert len(hash_fragment) == 3
+        # Check all hash characters are valid B256 characters
         for char in hash_fragment:
-            assert char in "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789~_@"
+            assert char in "".join(HashPos.B256)
 
 
 def test_hashline_with_start_line():
@@ -35,18 +36,18 @@ def test_hashline_with_start_line():
 
     lines = result.splitlines()
     assert len(lines) == 2
-    # Check format is {4-char-hash}::content (new HashPos format)
+    # Check format is {3-char-hash}::content (HashPos format)
     # Note: start_line parameter is ignored by HashPos but kept for compatibility
     for line in lines:
-        # Format should be "{4-char-hash}::content"
+        # Format should be "{3-char-hash}::content"
         assert "::" in line
         # Extract hash fragment (everything before "::")
         hash_fragment = line.split("::", 1)[0]
-        # Check hash fragment is 4 characters
-        assert len(hash_fragment) == 4
-        # Check all hash characters are valid base64 (A-Z, a-z, 0-9, ~, _, @)
+        # Check hash fragment is 3 characters
+        assert len(hash_fragment) == 3
+        # Check all hash characters are valid B256 characters
         for char in hash_fragment:
-            assert char in "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789~_@"
+            assert char in "".join(HashPos.B256)
     """Test hashline with empty string."""
     result = hashline("")
     assert result == ""
@@ -58,23 +59,23 @@ def test_hashline_single_line():
     result = hashline(text)
     lines = result.splitlines()
     assert len(lines) == 1
-    # Check format is {4-char-hash}::content (new HashPos format)
+    # Check format is {3-char-hash}::content (HashPos format)
     line = lines[0]
     assert "::" in line
     # Extract hash fragment (everything before "::")
     hash_fragment = line.split("::", 1)[0]
-    # Check hash fragment is 4 characters
-    assert len(hash_fragment) == 4
-    # Check all hash characters are valid base64 (A-Z, a-z, 0-9, ~, _, @)
+    # Check hash fragment is 3 characters
+    assert len(hash_fragment) == 3
+    # Check all hash characters are valid B256 characters
     for char in hash_fragment:
-        assert char in "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789~_@"
+        assert char in "".join(HashPos.B256)
 
 
 def test_hashline_preserves_newlines():
     """Test that hashline preserves newline characters."""
     text = "Line 1\nLine 2\n"
     result = hashline(text)
-    # HashPos format: {4-char-hash}::content on each line
+    # HashPos format: {3-char-hash}::content on each line
     # The result should have hashes on each line but no trailing newline
     lines = result.splitlines()
     assert len(lines) == 2
@@ -83,18 +84,18 @@ def test_hashline_preserves_newlines():
         assert "::" in line
         # Extract hash fragment (everything before "::")
         hash_fragment = line.split("::", 1)[0]
-        assert len(hash_fragment) == 4
-        # Check all hash characters are valid base64 (A-Z, a-z, 0-9, ~, _, @)
+        assert len(hash_fragment) == 3
+        # Check all hash characters are valid B256 characters
         for char in hash_fragment:
-            assert char in "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789~_@"
+            assert char in "".join(HashPos.B256)
     # HashPos doesn't preserve trailing newlines in the formatted output
     # The splitlines() above verifies we have the right number of lines
 
 
 def test_strip_hashline_basic():
     """Test basic strip_hashline functionality."""
-    # Create a hashline-formatted text with correct HashPos format: {4-char-hash}::content
-    text = "abcd::Hello\nefgh::World\nijkl::Test"
+    # Create a hashline-formatted text with correct HashPos format: {3-char-hash}::content
+    text = "abc::Hello\nefg::World\nijk::Test"
     stripped = strip_hashline(text)
     assert stripped == "Hello\nWorld\nTest"
 
@@ -103,24 +104,24 @@ def test_strip_hashline_with_negative_line_numbers():
     """Test strip_hashline with negative line numbers."""
     # HashPos format doesn't support negative line numbers in the prefix
     # Test with standard HashPos format
-    text = "abcd::Hello\nefgh::World\nijkl::Test"
+    text = "abc::Hello\nefg::World\nijk::Test"
     stripped = strip_hashline(text)
     assert stripped == "Hello\nWorld\nTest"
 
 
 def test_strip_hashline_mixed_lines():
     """Test strip_hashline with mixed hashline and non-hashline lines."""
-    # HashPos format: {4-char-hash}::content
+    # HashPos format: {3-char-hash}::content
     # Plain lines without hashes should be left unchanged
-    text = "abcd::Hello\nPlain line\nefgh::World"
+    text = "abc::Hello\nPlain line\nefg::World"
     stripped = strip_hashline(text)
     assert stripped == "Hello\nPlain line\nWorld"
 
 
 def test_strip_hashline_preserves_newlines():
     """Test that strip_hashline preserves newline characters."""
-    # HashPos format: {4-char-hash}::content
-    text = "abcd::Line 1\nefgh::Line 2\n"
+    # HashPos format: {3-char-hash}::content
+    text = "abc::Line 1\nefg::Line 2\n"
     stripped = strip_hashline(text)
     # strip_hashline should preserve newlines
     assert stripped == "Line 1\nLine 2\n"
@@ -154,14 +155,14 @@ def test_hashline_different_inputs():
     result1 = hashline(text1)
     result2 = hashline(text2)
 
-    # HashPos format: [{4-char-hash}]content
+    # HashPos format: {3-char-hash}::content
     # Extract hash from each line (there's only one line for single-line inputs)
     lines1 = result1.splitlines()
     lines2 = result2.splitlines()
 
-    # Get the hash from each line (format: [hash]content)
-    hash1 = lines1[0][1:5] if lines1 else ""  # Extract 4-char hash
-    hash2 = lines2[0][1:5] if lines2 else ""  # Extract 4-char hash
+    # Get the hash from each line (format: hash::content)
+    hash1 = lines1[0][:3] if lines1 else ""  # Extract 3-char hash
+    hash2 = lines2[0][:3] if lines2 else ""  # Extract 3-char hash
 
     # Hashes should be different (very high probability)
     assert hash1 != hash2
@@ -169,15 +170,15 @@ def test_hashline_different_inputs():
 
 def test_parse_hashline():
     """Test parse_hashline function."""
-    # Test basic parsing (HashPos format: {4-char-hash})
-    hash_fragment, line_num_str, line_num = parse_hashline("abcd")
-    assert hash_fragment == "abcd"
+    # Test basic parsing (HashPos format: {3-char-hash})
+    hash_fragment, line_num_str, line_num = parse_hashline("abc")
+    assert hash_fragment == "abc"
     assert line_num_str is None  # HashPos doesn't include line numbers
     assert line_num is None
 
     # Test with content after hash
-    hash_fragment, line_num_str, line_num = parse_hashline("efgh::Hello World")
-    assert hash_fragment == "efgh"
+    hash_fragment, line_num_str, line_num = parse_hashline("efg::Hello World")
+    assert hash_fragment == "efg"
     assert line_num_str is None
     assert line_num is None
 
diff --git a/tests/basic/test_main.py b/tests/basic/test_main.py
index 9daa7b3d94f..082877e2cb9 100644
--- a/tests/basic/test_main.py
+++ b/tests/basic/test_main.py
@@ -162,7 +162,7 @@ def test_main_copy_paste_model_overrides(dummy_io, git_temp_dir):
 
 
 def test_main_copy_paste_flag_sets_mode(dummy_io, git_temp_dir, mocker):
-    mock_watcher = mocker.patch("cecli.main.ClipboardWatcher")
+    mock_watcher = mocker.patch("cecli.helpers.copypaste.ClipboardWatcher")
     mock_watcher.return_value = MagicMock()
     coder = main(
         ["--no-git", "--exit", "--yes-always", "--copy-paste"], **dummy_io, return_coder=True
@@ -346,12 +346,17 @@ def test_encodings_arg(dummy_io, git_temp_dir, mocker):
     MockCoder = mocker.patch("cecli.coders.Coder.create")
     mock_coder_instance = MockCoder.return_value
     mock_coder_instance._autosave_future = mock_autosave_future()
-    MockSend = mocker.patch("cecli.main.InputOutput")
+    MockSend = mocker.patch("cecli.io.InputOutput")
 
     def side_effect(*args, **kwargs):
         assert kwargs["encoding"] == "iso-8859-15"
         mock_io = MagicMock()
         mock_io.confirm_ask = AsyncMock(return_value=True)
+        mock_io = MagicMock()
+        mock_io.confirm_ask = AsyncMock(return_value=True)
+        mock_io.offer_url = AsyncMock()
+        mock_io.tool_output = MagicMock()
+        mock_io.tool_error = MagicMock()
         return mock_io
 
     MockSend.side_effect = side_effect
@@ -359,9 +364,10 @@ def side_effect(*args, **kwargs):
 
 
 def test_main_exit_calls_version_check(dummy_io, git_temp_dir, mocker):
-    mock_check_version = mocker.patch("cecli.main.check_version")
-    mock_input_output = mocker.patch("cecli.main.InputOutput")
+    mock_check_version = mocker.patch("cecli.versioncheck.check_version")
+    mock_input_output = mocker.patch("cecli.io.InputOutput")
     mock_input_output.return_value.confirm_ask = AsyncMock(return_value=True)
+    mock_input_output.return_value.offer_url = AsyncMock()
     main(["--exit", "--check-update"], **dummy_io)
     mock_check_version.assert_called_once()
     mock_input_output.assert_called_once()
@@ -369,7 +375,7 @@ def test_main_exit_calls_version_check(dummy_io, git_temp_dir, mocker):
 
 def test_main_message_adds_to_input_history(dummy_io, mocker):
     mocker.patch("cecli.coders.base_coder.Coder.run")
-    MockInputOutput = mocker.patch("cecli.main.InputOutput", autospec=True)
+    MockInputOutput = mocker.patch("cecli.io.InputOutput", autospec=True)
     test_message = "test message"
     mock_io_instance = MockInputOutput.return_value
     mock_io_instance.pretty = True
@@ -379,7 +385,7 @@ def test_main_message_adds_to_input_history(dummy_io, mocker):
 
 def test_yes_always(dummy_io, mocker):
     mocker.patch("cecli.coders.base_coder.Coder.run")
-    MockInputOutput = mocker.patch("cecli.main.InputOutput", autospec=True)
+    MockInputOutput = mocker.patch("cecli.io.InputOutput", autospec=True)
     test_message = "test message"
     MockInputOutput.return_value.pretty = True
     main(["--yes-always", "--message", test_message])
@@ -389,7 +395,7 @@ def test_yes_always(dummy_io, mocker):
 
 def test_default_of_yes_all_is_none(dummy_io, mocker):
     mocker.patch("cecli.coders.base_coder.Coder.run")
-    MockInputOutput = mocker.patch("cecli.main.InputOutput", autospec=True)
+    MockInputOutput = mocker.patch("cecli.io.InputOutput", autospec=True)
     test_message = "test message"
     MockInputOutput.return_value.pretty = True
     main(["--message", test_message])
@@ -403,8 +409,9 @@ def test_default_of_yes_all_is_none(dummy_io, mocker):
     ids=["dark_mode", "light_mode"],
 )
 def test_mode_sets_code_theme(mode_flag, expected_theme, dummy_io, git_temp_dir, mocker):
-    MockInputOutput = mocker.patch("cecli.main.InputOutput")
+    MockInputOutput = mocker.patch("cecli.io.InputOutput")
     MockInputOutput.return_value.get_input.return_value = None
+    MockInputOutput.return_value.offer_url = AsyncMock()
     main([mode_flag, "--no-git", "--exit"], **dummy_io)
     MockInputOutput.assert_called_once()
     _, kwargs = MockInputOutput.call_args
@@ -429,9 +436,11 @@ def test_env_file_variables(
     env_file_path.write_text(env_content)
     is_dark_mode_test = check_attribute == "code_theme"
     if is_dark_mode_test:
-        MockInputOutput = mocker.patch("cecli.main.InputOutput")
+        MockInputOutput = mocker.patch("cecli.io.InputOutput")
         MockInputOutput.return_value.get_input.return_value = None
         MockInputOutput.return_value.get_input.confirm_ask = True
+        MockInputOutput.return_value.offer_url = AsyncMock()
+        MockInputOutput.return_value.get_input.confirm_ask = True
     args = ["--no-git", "--exit" if is_dark_mode_test else "--yes-always"]
     if use_flag:
         args.extend(["--env-file", str(env_file_path)])
@@ -1293,7 +1302,7 @@ def test_model_accepts_settings_attribute(dummy_io, git_temp_dir, mocker):
 )
 def test_stream_cache_warning(dummy_io, git_temp_dir, mocker, flags, should_warn):
     """Test warning shown only when both streaming and caching are enabled."""
-    MockInputOutput = mocker.patch("cecli.main.InputOutput", autospec=True)
+    MockInputOutput = mocker.patch("cecli.io.InputOutput", autospec=True)
     mock_io_instance = MockInputOutput.return_value
     mock_io_instance.pretty = True
     args = flags + ["--exit", "--yes-always"]
diff --git a/tests/basic/test_model_info_manager.py b/tests/basic/test_model_info_manager.py
index 8c62719d1a3..f7c802d298d 100644
--- a/tests/basic/test_model_info_manager.py
+++ b/tests/basic/test_model_info_manager.py
@@ -54,7 +54,7 @@ def test_lazy_loading_cache(self):
 
             # Verify cache was loaded
             self.assertTrue(self.manager._cache_loaded)
-            self.assertIsNotNone(self.manager.content)
+            self.assertIsNotNone(self.manager._raw_content)
             self.assertEqual(result, {"max_tokens": 4096})
 
             # Verify _update_cache was not called since cache exists and is valid
diff --git a/tests/helpers/observations/test_observation_service.py b/tests/helpers/observations/test_observation_service.py
index d51b8087da8..b456e06c2a6 100644
--- a/tests/helpers/observations/test_observation_service.py
+++ b/tests/helpers/observations/test_observation_service.py
@@ -93,8 +93,8 @@ async def test_compact_context_with_observations():
     # 2. compact (DONE)
     # 3. compact (CUR)
     # 4. compact (DIFFS)
-    mock_conv_manager.get_messages_dict.side_effect = [cur_messages, [], cur_messages, []]
-
+    # 5. compact (ALL)
+    mock_conv_manager.get_messages_dict.side_effect = [cur_messages, [], cur_messages, [], []]
     with patch(
         "cecli.coders.base_coder.ConversationService.get_manager", return_value=mock_conv_manager
     ):
@@ -154,8 +154,8 @@ async def test_compact_context_with_observations_integration():
     # 2. compact (DONE)
     # 3. compact (CUR)
     # 4. compact (DIFFS)
-    mock_conv_manager.get_messages_dict.side_effect = [cur_messages, [], cur_messages, []]
-
+    # 5. compact (ALL)
+    mock_conv_manager.get_messages_dict.side_effect = [cur_messages, [], cur_messages, [], []]
     with patch(
         "cecli.coders.base_coder.ConversationService.get_manager", return_value=mock_conv_manager
     ):
diff --git a/tests/integration/test_agent_mcp_management.py b/tests/integration/test_agent_mcp_management.py
new file mode 100644
index 00000000000..efabe6e6abb
--- /dev/null
+++ b/tests/integration/test_agent_mcp_management.py
@@ -0,0 +1,132 @@
+"""Integration tests for agent and subagent MCP management."""
+
+from unittest.mock import AsyncMock, MagicMock, Mock, patch
+
+import pytest
+
+from cecli.coders.agent_coder import AgentCoder
+from cecli.coders.sub_agent_coder import SubAgentCoder
+from cecli.tools.resource_manager import Tool as ResourceManagerTool
+
+
+@pytest.fixture
+def mock_mcp_manager():
+    """Fixture for a mocked McpServerManager."""
+    manager = MagicMock()
+    manager.connected_servers = {}
+
+    # Mock servers
+    server1 = MagicMock()
+    server1.name = "test_server"
+    server1.config = {"enabled": True}
+
+    server2 = MagicMock()
+    server2.name = "sub_test_server"
+    server2.config = {"enabled": True}
+
+    manager.servers = [server1, server2]
+
+    def get_server_side_effect(name):
+        if name == "test_server":
+            return server1
+        if name == "sub_test_server":
+            return server2
+        return None
+
+    manager.get_server.side_effect = get_server_side_effect
+
+    async def connect(server_name):
+        manager.connected_servers[server_name] = "connected"
+        return True, False  # (did_connect, interrupted)
+
+    async def disconnect(server_name):
+        if server_name in manager.connected_servers:
+            del manager.connected_servers[server_name]
+            return True, False
+        return False, False
+
+    manager.connect_server = AsyncMock(side_effect=connect)
+    manager.disconnect_server = AsyncMock(side_effect=disconnect)
+    manager.add_server = AsyncMock()
+    manager.add_server = AsyncMock()
+    manager.add_server = AsyncMock()
+    manager.add_server = AsyncMock()
+
+    return manager
+
+
+@pytest.fixture
+def agent_coder(mock_mcp_manager):
+    """Fixture for an AgentCoder with a mocked MCP manager."""
+    with patch("cecli.coders.agent_coder.McpServerManager", return_value=mock_mcp_manager):
+        coder = AgentCoder(
+            main_model=MagicMock(),
+            io=MagicMock(),
+        )
+        coder.mcp_manager = mock_mcp_manager
+        coder.original_kwargs = {}
+        coder.coroutines = Mock()
+
+        async def mock_interruptible(coro, event):
+            return await coro, False
+
+        coder.coroutines.interruptible.side_effect = mock_interruptible
+        coder.interrupt_event = Mock()
+        return coder
+
+
+@pytest.fixture
+async def sub_agent_coder(agent_coder):
+    """Fixture for a SubAgentCoder."""
+    # Fix: Use create() class method instead of direct instantiation
+    sub_agent = await SubAgentCoder.create(from_coder=agent_coder)
+    # Ensure sub_agent has the required mocks for tools
+    sub_agent.coroutines = agent_coder.coroutines
+    sub_agent.interrupt_event = agent_coder.interrupt_event
+    return sub_agent
+
+
+@pytest.mark.asyncio
+async def test_agent_can_load_mcp_server(agent_coder, mock_mcp_manager):
+    """Verify an agent can load an MCP server."""
+    server_name = "test_server"
+
+    await ResourceManagerTool.execute(agent_coder, load_mcp=[server_name])
+
+    mock_mcp_manager.connect_server.assert_called_once_with(server_name)
+    assert server_name in mock_mcp_manager.connected_servers
+
+
+@pytest.mark.asyncio
+async def test_agent_can_remove_mcp_server(agent_coder, mock_mcp_manager):
+    """Verify an agent can remove an MCP server."""
+    server_name = "test_server"
+    mock_mcp_manager.connected_servers[server_name] = "connected"
+
+    await ResourceManagerTool.execute(agent_coder, remove_mcp=[server_name])
+
+    mock_mcp_manager.disconnect_server.assert_called_once_with(server_name)
+    assert server_name not in mock_mcp_manager.connected_servers
+
+
+@pytest.mark.asyncio
+async def test_sub_agent_can_load_mcp_server(sub_agent_coder, mock_mcp_manager):
+    """Verify a subagent can load an MCP server."""
+    server_name = "sub_test_server"
+
+    await ResourceManagerTool.execute(sub_agent_coder, load_mcp=[server_name])
+
+    mock_mcp_manager.connect_server.assert_called_once_with(server_name)
+    assert server_name in mock_mcp_manager.connected_servers
+
+
+@pytest.mark.asyncio
+async def test_sub_agent_can_remove_mcp_server(sub_agent_coder, mock_mcp_manager):
+    """Verify a subagent can remove an MCP server."""
+    server_name = "sub_test_server"
+    mock_mcp_manager.connected_servers[server_name] = "connected"
+
+    await ResourceManagerTool.execute(sub_agent_coder, remove_mcp=[server_name])
+
+    mock_mcp_manager.disconnect_server.assert_called_once_with(server_name)
+    assert server_name not in mock_mcp_manager.connected_servers
diff --git a/tests/integration/test_mcp_management.py b/tests/integration/test_mcp_management.py
new file mode 100644
index 00000000000..003b34c6bf4
--- /dev/null
+++ b/tests/integration/test_mcp_management.py
@@ -0,0 +1,94 @@
+"""Integration tests for MCP management tools."""
+
+from unittest.mock import AsyncMock, MagicMock
+
+import pytest
+
+from cecli.tools.resource_manager import Tool as ResourceManagerTool
+
+
+class CoderMock:
+    """Mock Coder object for integration testing."""
+
+    def __init__(self):
+        self.mcp_manager = MagicMock()
+        self.mcp_manager.servers = []
+        self.mcp_manager.connected_servers = {}
+        self.mcp_manager.get_server.return_value = None
+        self.mcp_manager.connect_server = AsyncMock(return_value=(True, False))
+        self.mcp_manager.disconnect_server = AsyncMock(return_value=(True, False))
+        self.coroutines = MagicMock()
+        self.interrupt_event = MagicMock()
+        self.registered_servers = {"included": set(), "excluded": set()}
+        self.io = MagicMock()
+
+    async def mock_interruptible(self, coro, event):
+        """Mock interruptible that just executes the coroutine."""
+        return await coro
+
+    def add_server(self, name, enabled=True):
+        """Add a mock server to the manager."""
+        server = MagicMock()
+        server.name = name
+        server.config = {"enabled": enabled}
+        self.mcp_manager.servers.append(server)
+        original_get_server = self.mcp_manager.get_server.side_effect
+
+        def get_server_side_effect(server_name):
+            if server_name == name:
+                return server
+            if original_get_server:
+                return original_get_server(server_name)
+            return None
+
+        self.mcp_manager.get_server.side_effect = get_server_side_effect
+
+
+@pytest.fixture
+def coder():
+    """Provide a mock coder for integration testing."""
+    return CoderMock()
+
+
+@pytest.mark.asyncio
+async def test_integration_load_and_remove_server(coder):
+    """Test loading and then removing a server."""
+    coder.add_server("integration-test-server")
+    coder.coroutines.interruptible = coder.mock_interruptible
+
+    # Load the server
+    load_result = await ResourceManagerTool.execute(coder, load_mcp=["integration-test-server"])
+    assert "Loaded server: integration-test-server" in load_result
+
+    # Mock the connected server for the remove tool
+    coder.mcp_manager.connected_servers = {"integration-test-server": coder.mcp_manager.servers[0]}
+
+    # Remove the server
+    remove_result = await ResourceManagerTool.execute(coder, remove_mcp=["integration-test-server"])
+    assert "Removed server: integration-test-server" in remove_result
+
+
+@pytest.mark.asyncio
+async def test_integration_wildcard_load_and_remove(coder):
+    """Test loading and removing all servers with a wildcard."""
+    coder.add_server("server1")
+    coder.add_server("server2")
+    coder.add_server("server3", enabled=False)
+    coder.coroutines.interruptible = coder.mock_interruptible
+
+    # Load all enabled servers
+    load_result = await ResourceManagerTool.execute(coder, load_mcp=["*"])
+    assert "Loaded server: server1" in load_result
+    assert "Loaded server: server2" in load_result
+    # Non-enabled servers are filtered out silently by wildcard expansion
+
+    # Mock the connected servers for the remove tool
+    coder.mcp_manager.connected_servers = {
+        "server1": coder.mcp_manager.servers[0],
+        "server2": coder.mcp_manager.servers[1],
+    }
+
+    # Remove all connected servers
+    remove_result = await ResourceManagerTool.execute(coder, remove_mcp=["*"])
+    assert "Removed server: server1" in remove_result
+    assert "Removed server: server2" in remove_result
diff --git a/tests/tools/test_registry.py b/tests/tools/test_registry.py
index 7f540e38bde..5cbddf7d529 100644
--- a/tests/tools/test_registry.py
+++ b/tests/tools/test_registry.py
@@ -28,17 +28,17 @@ def test_registry_initialization(self):
         assert len(tools) > 0, "Registry should have tools after initialization"
 
         # Check that essential tools are registered
-        essential_tools = {"contextmanager", "edittext", "yield"}
+        essential_tools = {"resourcemanager", "edittext", "yield"}
         for tool in essential_tools:
             assert tool in tools, f"Essential tool {tool} should be registered"
 
     def test_get_tool(self):
         """Test getting individual tools by name"""
         # Get existing tool
-        tool_class = ToolRegistry.get_tool("contextmanager")
-        assert tool_class is not None, "Should get contextmanager tool"
+        tool_class = ToolRegistry.get_tool("resourcemanager")
+        assert tool_class is not None, "Should get resourcemanager tool"
         assert hasattr(tool_class, "NORM_NAME"), "Tool class should have NORM_NAME"
-        assert tool_class.NORM_NAME == "contextmanager", "Tool name should match"
+        assert tool_class.NORM_NAME == "resourcemanager", "Tool name should match"
 
         # Get non-existent tool
         non_existent = ToolRegistry.get_tool("nonexistenttool")
@@ -52,18 +52,18 @@ def test_build_registry_empty_config(self):
         assert len(registry) > 0, "Should return tools with empty config"
 
         # Essential tools should always be included
-        assert "contextmanager" in registry, "Essential tool should be included"
+        assert "resourcemanager" in registry, "Essential tool should be included"
         assert "edittext" in registry, "Essential tool should be included"
         assert "yield" in registry, "Essential tool should be included"
 
     def test_build_registry_with_includelist(self):
         """Test filtering with tools_includelist"""
-        config = {"tools_includelist": ["contextmanager", "edittext"]}
+        config = {"tools_includelist": ["resourcemanager", "edittext"]}
         registry = ToolRegistry.build_registry(config)
 
         # Should only include tools from includelist, plus essential tools
         assert len(registry) == 3, "Should include 2 from list + 1 essential"
-        assert "contextmanager" in registry
+        assert "resourcemanager" in registry
         assert "edittext" in registry
         assert "yield" in registry  # Essential
         assert "command" not in registry, "Should not include tools not in includelist"
@@ -76,15 +76,15 @@ def test_build_registry_with_excludelist(self):
         # Should exclude specified tools (except essentials)
         assert "command" not in registry, "Should exclude command"
         assert "commandinteractive" not in registry, "Should exclude commandinteractive"
-        assert "contextmanager" in registry, "Essential tool should still be included"
+        assert "resourcemanager" in registry, "Essential tool should still be included"
 
     def test_build_registry_exclude_essential(self):
         """Test that essential tools cannot be excluded"""
-        config = {"tools_excludelist": ["contextmanager", "edittext", "finished", "command"]}
+        config = {"tools_excludelist": ["resourcemanager", "edittext", "finished", "command"]}
         registry = ToolRegistry.build_registry(config)
 
         # Essential tools should still be included despite excludelist
-        assert "contextmanager" in registry, "Essential tool cannot be excluded"
+        assert "resourcemanager" in registry, "Essential tool cannot be excluded"
         assert "edittext" in registry, "Essential tool cannot be excluded"
         assert "yield" in registry, "Essential tool cannot be excluded"
         assert "command" not in registry, "Non-essential tool should be excluded"
@@ -92,14 +92,14 @@ def test_build_registry_exclude_essential(self):
     def test_build_registry_combined_filters(self):
         """Test combined filtering with includelist and excludelist"""
         config = {
-            "tools_includelist": ["contextmanager", "edittext", "command"],
+            "tools_includelist": ["resourcemanager", "edittext", "command"],
             "tools_excludelist": ["commandinteractive"],
         }
         registry = ToolRegistry.build_registry(config)
 
         # Should respect all filters
         assert len(registry) == 4, "Should include exactly 4 tools (3 from list + yield)"
-        assert "contextmanager" in registry
+        assert "resourcemanager" in registry
         assert "edittext" in registry
         assert "yield" in registry
         assert "command" in registry
@@ -107,35 +107,35 @@ def test_build_registry_combined_filters(self):
 
     def test_get_filtered_tools(self):
         """Test get_filtered_tools method"""
-        config = {"tools_includelist": ["contextmanager", "edittext"]}
+        config = {"tools_includelist": ["resourcemanager", "edittext"]}
         ToolRegistry.build_registry(config)
         tool_names = ToolRegistry.get_registered_tools()
 
         # Should return list of tool names
         assert isinstance(tool_names, list)
-        # Should include contextmanager, edittext, and finished (essential)
+        # Should include resourcemanager, edittext, and finished (essential)
         assert len(tool_names) == 3
-        assert "contextmanager" in tool_names
+        assert "resourcemanager" in tool_names
         assert "edittext" in tool_names
         assert "yield" in tool_names  # Essential tool always included
 
     def test_legacy_config_names(self):
         """Test backward compatibility with legacy config names (whitelist/blacklist)"""
         config = {
-            "tools_whitelist": ["contextmanager", "edittext"],
+            "tools_whitelist": ["resourcemanager", "edittext"],
             "tools_blacklist": ["command"],
         }
         registry = ToolRegistry.build_registry(config)
 
         # Should work with legacy names
-        assert "contextmanager" in registry
+        assert "resourcemanager" in registry
         assert "edittext" in registry
         assert "command" not in registry
 
     def test_config_precedence(self):
         """Test that new config names take precedence over legacy names"""
         config = {
-            "tools_includelist": ["contextmanager"],
+            "tools_includelist": ["resourcemanager"],
             "tools_whitelist": ["command"],  # Should be ignored
             "tools_excludelist": ["commandinteractive"],
             "tools_blacklist": ["finished"],  # Should be ignored for essential tool
@@ -143,7 +143,7 @@ def test_config_precedence(self):
         registry = ToolRegistry.build_registry(config)
 
         # New names should take precedence
-        assert "contextmanager" in registry, "Should use tools_includelist"
+        assert "resourcemanager" in registry, "Should use tools_includelist"
         assert (
             "command" not in registry
         ), "Should not use tools_whitelist when tools_includelist present"
@@ -152,7 +152,7 @@ def test_config_precedence(self):
 
     def test_registry_consistency(self):
         """Test that registry methods return consistent results"""
-        config = {"tools_includelist": ["contextmanager", "edittext"]}
+        config = {"tools_includelist": ["resourcemanager", "edittext"]}
 
         # build_registry should return consistent results
         registry = ToolRegistry.build_registry(config)
@@ -163,21 +163,24 @@ def test_registry_consistency(self):
         ), "Methods should return consistent results"
         assert len(registry) == len(filtered_names), "Methods should return consistent counts"
 
-    def test_skill_tool_detection(self):
-        """Test that skill tools are correctly identified"""
-        # Get the actual tool classes to verify
-        loadskill_tool = ToolRegistry.get_tool("loadskill")
-        removeskill_tool = ToolRegistry.get_tool("removeskill")
-
-        # These should exist in the registry
-        assert loadskill_tool is not None, "loadskill tool should be registered"
-        assert removeskill_tool is not None, "removeskill tool should be registered"
-
-        # Verify they have the correct NORM_NAME
-        if loadskill_tool:
-            assert loadskill_tool.NORM_NAME == "loadskill"
-        if removeskill_tool:
-            assert removeskill_tool.NORM_NAME == "removeskill"
+    def test_skill_and_mcp_tools_in_context_manager(self):
+        """Test that skill/MCP functionality is now part of context_manager."""
+        # The individual load_skill, remove_skill, load_mcp, remove_mcp tools
+        # have been merged into context_manager
+        ctx_tool = ToolRegistry.get_tool("resourcemanager")
+        assert ctx_tool is not None, "resourcemanager tool should be registered"
+        assert ctx_tool.NORM_NAME == "resourcemanager"
+
+        # Verify the merged tools no longer exist as separate entries
+        assert ToolRegistry.get_tool("loadskill") is None
+        assert ToolRegistry.get_tool("removeskill") is None
+
+        # Verify context_manager has the new schema parameters
+        params = ctx_tool.SCHEMA["function"]["parameters"]["properties"]
+        assert "load_skill" in params, "context_manager should have load_skill param"
+        assert "remove_skill" in params, "context_manager should have remove_skill param"
+        assert "load_mcp" in params, "context_manager should have load_mcp param"
+        assert "remove_mcp" in params, "context_manager should have remove_mcp param"
 
 
 if __name__ == "__main__":
diff --git a/tests/tools/test_remove_mcp_tool.py b/tests/tools/test_remove_mcp_tool.py
new file mode 100644
index 00000000000..ddd2e6e6045
--- /dev/null
+++ b/tests/tools/test_remove_mcp_tool.py
@@ -0,0 +1,157 @@
+"""Unit tests for RemoveMcpTool.execute."""
+
+from unittest.mock import AsyncMock, Mock, patch
+
+import pytest
+
+from cecli.tools.resource_manager import Tool as ResourceManagerTool
+
+
+class DummyIO:
+    """Mock IO object for testing."""
+
+    def __init__(self):
+        self.tool_error = Mock()
+        self.tool_warning = Mock()
+        self.tool_output = Mock()
+        self.interrupt_event = Mock()
+
+
+class DummyCoder:
+    """Mock Coder object for testing."""
+
+    def __init__(self):
+        self.io = DummyIO()
+        self.mcp_manager = Mock()
+        self.mcp_manager.servers = []
+        self.mcp_manager.connected_servers = {}
+        self.coroutines = Mock()
+        self.coroutines.interruptible = AsyncMock()
+
+        self.interrupt_event = Mock()
+        self.registered_servers = {"included": set(), "excluded": set()}
+        self.agent_config = {"include_context_blocks": {"servers"}, "exclude_context_blocks": set()}
+
+
+@pytest.fixture
+def coder():
+    """Provide a dummy coder for testing."""
+    return DummyCoder()
+
+
+@pytest.fixture
+def mock_server():
+    """Provide a mock MCP server."""
+    server = Mock()
+    server.name = "test-server"
+    return server
+
+
+class TestRemoveMcpTool:
+    """Test cases for RemoveMcpTool."""
+
+    @pytest.fixture(autouse=True)
+    def _patch_global_exclusion(self):
+        """Patch is_server_globally_excluded to prevent test isolation issues with AgentService."""
+        with patch(
+            "cecli.tools.resource_manager.is_server_globally_excluded",
+            return_value=True,
+        ):
+            yield
+
+    @pytest.mark.asyncio
+    async def test_no_configured_servers(self, coder):
+        """Test when no MCP servers are configured at all."""
+        coder.mcp_manager.servers = []
+        result = await ResourceManagerTool.execute(coder, remove_mcp=["test"])
+        assert result == "No MCP servers are configured."
+
+    @pytest.mark.asyncio
+    async def test_server_not_found(self, coder, mock_server):
+        """Test when requested server doesn't exist."""
+        coder.mcp_manager.servers = [mock_server]
+        coder.mcp_manager.connected_servers = {"existing": "server"}
+        coder.mcp_manager.get_server.return_value = None
+        result = await ResourceManagerTool.execute(coder, remove_mcp=["nonexistent"])
+        assert "MCP server nonexistent does not exist." in result
+
+    @pytest.mark.asyncio
+    async def test_all_servers_not_loaded(self, coder, mock_server):
+        """Test when multiple servers exist but are not loaded."""
+        coder.mcp_manager.servers = [mock_server]
+        coder.mcp_manager.connected_servers = {}
+        coder.mcp_manager.get_server.return_value = mock_server
+        result = await ResourceManagerTool.execute(coder, remove_mcp=["test-server"])
+        assert "Server test-server is not currently connected." in result
+
+    @pytest.mark.asyncio
+    async def test_successful_removal(self, coder, mock_server):
+        """Test successful server removal."""
+        coder.mcp_manager.servers = [mock_server]
+        coder.mcp_manager.connected_servers = {"test-server": mock_server}
+        coder.mcp_manager.get_server.return_value = mock_server
+        coder.coroutines.interruptible.return_value = (True, False)
+        result = await ResourceManagerTool.execute(coder, remove_mcp=["test-server"])
+        assert "Removed server: test-server" in result
+
+    @pytest.mark.asyncio
+    async def test_removal_interrupted(self, coder, mock_server):
+        """Test when removal is interrupted."""
+        coder.mcp_manager.servers = [mock_server]
+        coder.mcp_manager.connected_servers = {"test-server": mock_server}
+        coder.mcp_manager.get_server.return_value = mock_server
+        coder.coroutines.interruptible.return_value = (False, True)
+        result = await ResourceManagerTool.execute(coder, remove_mcp=["test-server"])
+        assert "Interrupted: test-server" in result
+
+    @pytest.mark.asyncio
+    async def test_removal_failed(self, coder, mock_server):
+        """Test when removal fails."""
+        coder.mcp_manager.servers = [mock_server]
+        coder.mcp_manager.connected_servers = {"test-server": mock_server}
+        coder.mcp_manager.get_server.return_value = mock_server
+        coder.coroutines.interruptible.return_value = (False, False)
+        result = await ResourceManagerTool.execute(coder, remove_mcp=["test-server"])
+        assert "Unable to remove server: test-server" in result
+
+    @pytest.mark.asyncio
+    async def test_remove_all_servers(self, coder):
+        """Test removing all servers with '*' wildcard."""
+        server1 = Mock()
+        server1.name = "server1"
+        server2 = Mock()
+        server2.name = "server2"
+        coder.mcp_manager.servers = [server1, server2]
+        coder.mcp_manager.connected_servers = {"server1": server1, "server2": server2}
+        coder.mcp_manager.get_server.side_effect = lambda name: next(
+            (s for s in [server1, server2] if s.name == name), None
+        )
+        coder.coroutines.interruptible.return_value = (True, False)
+        result = await ResourceManagerTool.execute(coder, remove_mcp=["*"])
+        assert "Removed server: server1" in result
+        assert "Removed server: server2" in result
+
+    @pytest.mark.asyncio
+    async def test_mixed_results(self, coder):
+        """Test mixed success/failure results."""
+        server1 = Mock()
+        server1.name = "server1"
+        server2 = Mock()
+        server2.name = "server2"
+        coder.mcp_manager.servers = [server1, server2]
+        coder.mcp_manager.connected_servers = {"server1": server1, "server2": server2}
+        coder.mcp_manager.get_server.side_effect = lambda name: next(
+            (s for s in [server1, server2] if s.name == name), None
+        )
+        call_count = 0
+
+        async def mock_interruptible_func(*args, **kwargs):
+            nonlocal call_count
+            result = (True, False) if call_count == 0 else (False, False)
+            call_count += 1
+            return result
+
+        coder.coroutines.interruptible.side_effect = mock_interruptible_func
+        result = await ResourceManagerTool.execute(coder, remove_mcp=["server1", "server2"])
+        assert "Removed server: server1" in result
+        assert "Unable to remove server: server2" in result
diff --git a/tests/tools/test_tools_load_mcp_tool.py b/tests/tools/test_tools_load_mcp_tool.py
new file mode 100644
index 00000000000..d5a40be0dd8
--- /dev/null
+++ b/tests/tools/test_tools_load_mcp_tool.py
@@ -0,0 +1,218 @@
+"""Unit tests for LoadMcpTool.execute."""
+
+from unittest.mock import AsyncMock, Mock
+
+import pytest
+
+from cecli.tools.resource_manager import Tool as ResourceManagerTool
+
+
+class DummyIO:
+    """Mock IO object for testing."""
+
+    def __init__(self):
+        self.tool_error = Mock()
+        self.tool_warning = Mock()
+        self.tool_output = Mock()
+        self.interrupt_event = Mock()
+
+
+class DummyCoder:
+    """Mock Coder object for testing."""
+
+    def __init__(self):
+        self.io = DummyIO()
+        self.mcp_manager = Mock()
+        self.mcp_manager.servers = []
+        self.mcp_manager.connected_servers = {}
+        self.coroutines = Mock()
+        self.coroutines.interruptible = AsyncMock()
+        self.interrupt_event = Mock()
+        self.registered_servers = {"included": set(), "excluded": set()}
+        self.agent_config = {"include_context_blocks": {"servers"}, "exclude_context_blocks": set()}
+
+
+@pytest.fixture
+def coder():
+    """Provide a dummy coder for testing."""
+    return DummyCoder()
+
+
+@pytest.fixture
+def mock_server():
+    """Provide a mock MCP server."""
+    server = Mock()
+    server.name = "test-server"
+    server.config = {"enabled": True}
+    return server
+
+
+class TestLoadMcpTool:
+    """Test cases for LoadMcpTool."""
+
+    @pytest.mark.asyncio
+    async def test_no_mcp_servers_found(self, coder):
+        """Test when no MCP servers are configured."""
+        coder.mcp_manager.servers = []
+        result = await ResourceManagerTool.execute(coder, load_mcp=["test"])
+        assert result == "No MCP servers found, nothing to load."
+
+    @pytest.mark.asyncio
+    async def test_server_not_found(self, coder, mock_server):
+        """Test when requested server doesn't exist."""
+        coder.mcp_manager.servers = [mock_server]
+        coder.mcp_manager.get_server.return_value = None
+        result = await ResourceManagerTool.execute(coder, load_mcp=["nonexistent"])
+        assert "MCP server nonexistent does not exist." in result
+
+    @pytest.mark.asyncio
+    async def test_server_already_loaded(self, coder, mock_server):
+        """Test when server is already loaded."""
+        mock_server.name = "test-server"
+        coder.mcp_manager.servers = [mock_server]
+        coder.mcp_manager.connected_servers = {"test-server": mock_server}
+        coder.mcp_manager.get_server.return_value = mock_server
+        # Must return tuple (did_connect, interrupted)
+        coder.coroutines.interruptible.return_value = (True, False)
+        result = await ResourceManagerTool.execute(coder, load_mcp=["test-server"])
+        assert "Server already loaded: test-server" in result
+
+    @pytest.mark.asyncio
+    async def test_server_not_enabled_by_default(self, coder, mock_server):
+        """Test when server is not enabled by default."""
+        mock_server.config = {"enabled": False}
+        coder.mcp_manager.servers = [mock_server]
+        coder.mcp_manager.get_server.return_value = mock_server
+        await ResourceManagerTool.execute(coder, load_mcp=["*"])
+
+    @pytest.mark.asyncio
+    async def test_successful_load(self, coder, mock_server):
+        """Test successful server loading."""
+        coder.mcp_manager.servers = [mock_server]
+        coder.mcp_manager.connected_servers = {}
+        coder.mcp_manager.get_server.return_value = mock_server
+        coder.coroutines.interruptible.return_value = (True, False)
+        result = await ResourceManagerTool.execute(coder, load_mcp=["test-server"])
+        assert "Loaded server: test-server" in result
+
+    @pytest.mark.asyncio
+    async def test_load_interrupted(self, coder, mock_server):
+        """Test when loading is interrupted."""
+        coder.mcp_manager.servers = [mock_server]
+        coder.mcp_manager.connected_servers = {}
+        coder.mcp_manager.get_server.return_value = mock_server
+        coder.coroutines.interruptible.return_value = (False, True)
+        result = await ResourceManagerTool.execute(coder, load_mcp=["test-server"])
+        assert "Interrupted: test-server" in result
+
+    @pytest.mark.asyncio
+    async def test_load_failed(self, coder, mock_server):
+        """Test when loading fails."""
+        coder.mcp_manager.servers = [mock_server]
+        coder.mcp_manager.connected_servers = {}
+        coder.mcp_manager.get_server.return_value = mock_server
+        coder.coroutines.interruptible.return_value = (False, False)
+        result = await ResourceManagerTool.execute(coder, load_mcp=["test-server"])
+        assert "Unable to load server: test-server" in result
+
+    @pytest.mark.asyncio
+    async def test_load_all_servers(self, coder):
+        """Test loading all servers with '*' wildcard."""
+        server1 = Mock()
+        server1.name = "server1"
+        server1.config = {"enabled": True}
+        server2 = Mock()
+        server2.name = "server2"
+        server2.config = {"enabled": True}
+        coder.mcp_manager.servers = [server1, server2]
+        coder.mcp_manager.connected_servers = {}
+        coder.mcp_manager.get_server.side_effect = lambda name: next(
+            (s for s in [server1, server2] if s.name == name), None
+        )
+        coder.coroutines.interruptible.return_value = (True, False)
+        result = await ResourceManagerTool.execute(coder, load_mcp=["*"])
+        assert "Loaded server: server1" in result
+        assert "Loaded server: server2" in result
+
+    @pytest.mark.asyncio
+    async def test_mixed_results(self, coder):
+        """Test mixed success/failure results."""
+        server1 = Mock()
+        server1.name = "server1"
+        server1.config = {"enabled": True}
+        server2 = Mock()
+        server2.name = "server2"
+        server2.config = {"enabled": True}
+        coder.mcp_manager.servers = [server1, server2]
+        coder.mcp_manager.connected_servers = {}
+        coder.mcp_manager.get_server.side_effect = lambda name: next(
+            (s for s in [server1, server2] if s.name == name), None
+        )
+
+        async def mock_interruptible_func(*args, **kwargs):
+            # First call succeeds, second fails
+            if not hasattr(mock_interruptible_func, "call_count"):
+                mock_interruptible_func.call_count = 0
+            mock_interruptible_func.call_count += 1
+            if mock_interruptible_func.call_count == 1:
+                return (True, False)
+            else:
+                return (False, False)
+
+        coder.coroutines.interruptible.side_effect = mock_interruptible_func
+        result = await ResourceManagerTool.execute(coder, load_mcp=["server1", "server2"])
+        assert "Loaded server: server1" in result
+        assert "Unable to load server: server2" in result
+
+    @pytest.mark.asyncio
+    async def test_duplicate_iteration_bug_fix(self, coder, mock_server):
+        """Test that duplicate iteration bug is fixed - server already loaded only processed once."""
+        mock_server.name = "test-server"
+        coder.mcp_manager.servers = [mock_server]
+        # Server already connected
+        coder.mcp_manager.connected_servers = {"test-server": mock_server}
+        coder.mcp_manager.get_server.return_value = mock_server
+
+        result = await ResourceManagerTool.execute(coder, load_mcp=["test-server"])
+
+        # Should only report server already loaded once
+        assert result.count("Server already loaded: test-server") == 1
+        # connect_server should not have been called since it was already loaded
+        coder.mcp_manager.connect_server.assert_not_called()
+
+    @pytest.mark.asyncio
+    async def test_wildcard_with_duplicate_iteration_fix(self, coder):
+        """Test wildcard loading with duplicate iteration fix."""
+        server1 = Mock()
+        server1.name = "server1"
+        server1.config = {"enabled": True}
+        server2 = Mock()
+        server2.name = "server2"
+        server2.config = {"enabled": True}
+        coder.mcp_manager.servers = [server1, server2]
+        # server1 already loaded, server2 not loaded
+        coder.mcp_manager.connected_servers = {"server1": server1}
+        coder.mcp_manager.get_server.side_effect = lambda name: next(
+            (s for s in [server1, server2] if s.name == name), None
+        )
+        connect_calls = []
+
+        async def mock_connect_server(server_name):
+            connect_calls.append(server_name)
+            if server_name == "server2":
+                return True, False
+            return False, False
+
+        async def mock_interruptible(coro, event):
+            return await coro
+
+        from unittest.mock import AsyncMock
+
+        coder.mcp_manager.connect_server = AsyncMock(side_effect=mock_connect_server)
+        coder.coroutines.interruptible = AsyncMock(side_effect=mock_interruptible)
+        result = await ResourceManagerTool.execute(coder, load_mcp=["*"])
+
+        # Should only attempt to load server2 (server1 should be skipped)
+        # Wildcard expansion skips already-connected servers, so server1 is not reported
+        assert "Loaded server: server2" in result
+        assert connect_calls == ["server2"]  # Only server2 should have been connected
diff --git a/tests/unit/test_load_mcp.py b/tests/unit/test_load_mcp.py
new file mode 100644
index 00000000000..1f33511a4b3
--- /dev/null
+++ b/tests/unit/test_load_mcp.py
@@ -0,0 +1,281 @@
+"""Unit tests for LoadMcpTool.execute."""
+
+from unittest.mock import AsyncMock, MagicMock
+
+import pytest
+
+from cecli.tools.resource_manager import Tool as ResourceManagerTool
+
+
+class DummyIO:
+    """Mock IO object for testing."""
+
+    def __init__(self):
+        self.tool_error = MagicMock()
+        self.tool_warning = MagicMock()
+        self.tool_output = MagicMock()
+        self.interrupt_event = MagicMock()
+
+
+class DummyCoder:
+    """Mock Coder object for testing."""
+
+    def __init__(self):
+        self.io = DummyIO()
+        self.mcp_manager = MagicMock()
+        self.mcp_manager.servers = []
+        self.mcp_manager.connected_servers = {}
+        self.coroutines = MagicMock()
+        self.interrupt_event = MagicMock()
+        self.registered_servers = {"included": set(), "excluded": set()}
+        self.agent_config = {"include_context_blocks": {"servers"}, "exclude_context_blocks": set()}
+
+
+@pytest.fixture
+def coder():
+    """Provide a dummy coder for testing."""
+    return DummyCoder()
+
+
+@pytest.fixture
+def mock_server():
+    """Provide a mock MCP server."""
+    server = MagicMock()
+    server.name = "test-server"
+    server.config = {"enabled": True}
+    return server
+
+
+@pytest.mark.asyncio
+async def test_no_mcp_servers_found(coder):
+    """Test when no MCP servers are configured."""
+    coder.mcp_manager.servers = []
+    result = await ResourceManagerTool.execute(coder, load_mcp=["test"])
+    assert result == "No MCP servers found, nothing to load."
+
+
+@pytest.mark.asyncio
+async def test_server_not_found(coder, mock_server):
+    """Test when requested server doesn't exist."""
+    coder.mcp_manager.servers = [mock_server]
+    coder.mcp_manager.get_server.return_value = None
+    result = await ResourceManagerTool.execute(coder, load_mcp=["nonexistent"])
+    assert "MCP server nonexistent does not exist." in result
+
+
+@pytest.mark.asyncio
+async def test_server_already_loaded(coder, mock_server):
+    """Test when server is already loaded."""
+    mock_server.name = "test-server"
+    coder.mcp_manager.servers = [mock_server]
+    coder.mcp_manager.connected_servers = {"test-server": mock_server}
+    coder.mcp_manager.get_server.return_value = mock_server
+    # Set up connect_server as AsyncMock so assert_not_called works
+    coder.mcp_manager.connect_server = AsyncMock()
+
+    async def mock_interruptible(coro, event):
+        return await coro, False
+
+    coder.coroutines.interruptible = mock_interruptible
+    result = await ResourceManagerTool.execute(coder, load_mcp=["test-server"])
+    assert "Server already loaded: test-server" in result
+    # connect_server should not have been called since it was already loaded
+    coder.mcp_manager.connect_server.assert_not_called()
+
+
+@pytest.mark.asyncio
+async def test_server_not_enabled_by_default(coder, mock_server):
+    """Test when server is not enabled by default."""
+    mock_server.config = {"enabled": False}
+    coder.mcp_manager.servers = [mock_server]
+    coder.mcp_manager.connected_servers = {}
+    coder.mcp_manager.get_server.return_value = mock_server
+    coder.mcp_manager.connect_server = AsyncMock()
+    result = await ResourceManagerTool.execute(coder, load_mcp=["*"])
+    # Non-enabled servers are silently filtered by wildcard expansion
+    assert result == ""
+    coder.mcp_manager.connect_server.assert_not_called()
+
+
+@pytest.mark.asyncio
+async def test_successful_load(coder, mock_server):
+    """Test successful server loading."""
+    coder.mcp_manager.servers = [mock_server]
+    coder.mcp_manager.connected_servers = {}
+    coder.mcp_manager.get_server.return_value = mock_server
+
+    # Set up connect_server as AsyncMock that returns (True, False)
+    async def mock_connect_server(server_name):
+        return True, False
+
+    coder.mcp_manager.connect_server = mock_connect_server
+
+    # Mock interruptible to just execute the coroutine
+    async def mock_interruptible(coro, event):
+        return await coro, False
+
+    coder.coroutines.interruptible = mock_interruptible
+    result = await ResourceManagerTool.execute(coder, load_mcp=["test-server"])
+    assert "Loaded server: test-server" in result
+
+
+@pytest.mark.asyncio
+async def test_load_interrupted(coder, mock_server):
+    """Test when loading is interrupted."""
+    coder.mcp_manager.servers = [mock_server]
+    coder.mcp_manager.connected_servers = {}
+    coder.mcp_manager.get_server.return_value = mock_server
+
+    # Set up connect_server as AsyncMock
+    async def mock_connect_server(server_name):
+        return True, False
+
+    coder.mcp_manager.connect_server = mock_connect_server
+
+    # Mock interruptible to return interruption
+    async def mock_interruptible(coro, event):
+        return False, True
+
+    coder.coroutines.interruptible = mock_interruptible
+    result = await ResourceManagerTool.execute(coder, load_mcp=["test-server"])
+    assert "Interrupted: test-server" in result
+
+
+@pytest.mark.asyncio
+async def test_load_failed(coder, mock_server):
+    """Test when loading fails."""
+    coder.mcp_manager.servers = [mock_server]
+    coder.mcp_manager.connected_servers = {}
+    coder.mcp_manager.get_server.return_value = mock_server
+
+    # Set up connect_server as AsyncMock that returns failure
+    coder.mcp_manager.connect_server = AsyncMock(return_value=False)
+
+    # Mock interruptible to just execute the coroutine
+    async def mock_interruptible(coro, event):
+        return await coro, False
+
+    coder.coroutines.interruptible = mock_interruptible
+    result = await ResourceManagerTool.execute(coder, load_mcp=["test-server"])
+    assert "Unable to load server: test-server" in result
+
+
+@pytest.mark.asyncio
+async def test_load_all_servers(coder):
+    """Test loading all servers with '*' wildcard."""
+    server1 = MagicMock()
+    server1.name = "server1"
+    server1.config = {"enabled": True}
+    server2 = MagicMock()
+    server2.name = "server2"
+    server2.config = {"enabled": True}
+    coder.mcp_manager.servers = [server1, server2]
+    coder.mcp_manager.connected_servers = {}
+    coder.mcp_manager.get_server.side_effect = lambda name: next(
+        (s for s in [server1, server2] if s.name == name), None
+    )
+
+    # Set up connect_server as AsyncMock
+    async def mock_connect_server(server_name):
+        return True, False
+
+    coder.mcp_manager.connect_server = mock_connect_server
+
+    # Mock interruptible to just execute the coroutine
+    async def mock_interruptible(coro, event):
+        return await coro, False
+
+    coder.coroutines.interruptible = mock_interruptible
+    result = await ResourceManagerTool.execute(coder, load_mcp=["*"])
+    assert "Loaded server: server1" in result
+    assert "Loaded server: server2" in result
+
+
+@pytest.mark.asyncio
+async def test_mixed_results(coder):
+    """Test mixed success/failure results."""
+    server1 = MagicMock()
+    server1.name = "server1"
+    server1.config = {"enabled": True}
+    server2 = MagicMock()
+    server2.name = "server2"
+    server2.config = {"enabled": True}
+    coder.mcp_manager.servers = [server1, server2]
+    coder.mcp_manager.connected_servers = {}
+    coder.mcp_manager.get_server.side_effect = lambda name: next(
+        (s for s in [server1, server2] if s.name == name), None
+    )
+    # First call succeeds, second fails
+    call_count = 0
+
+    async def mock_connect_server(server_name):
+        nonlocal call_count
+        result = True if call_count == 0 else False
+        call_count += 1
+        return result
+
+    coder.mcp_manager.connect_server = mock_connect_server
+
+    # Mock interruptible to just execute the coroutine
+    async def mock_interruptible(coro, event):
+        return await coro, False
+
+    coder.coroutines.interruptible = mock_interruptible
+    result = await ResourceManagerTool.execute(coder, load_mcp=["server1", "server2"])
+    assert "Loaded server: server1" in result
+    assert "Unable to load server: server2" in result
+
+
+@pytest.mark.asyncio
+async def test_duplicate_iteration_bug_fix(coder, mock_server):
+    """Test that duplicate iteration bug is fixed - server already loaded only processed once."""
+    mock_server.name = "test-server"
+    coder.mcp_manager.servers = [mock_server]
+    # Server already connected
+    coder.mcp_manager.connected_servers = {"test-server": mock_server}
+    coder.mcp_manager.get_server.return_value = mock_server
+    # Set up connect_server as AsyncMock
+    coder.mcp_manager.connect_server = AsyncMock()
+    result = await ResourceManagerTool.execute(coder, load_mcp=["test-server"])
+    # Should only report server already loaded once
+    assert result.count("Server already loaded: test-server") == 1
+    # connect_server should not have been called since it was already loaded
+    coder.mcp_manager.connect_server.assert_not_called()
+
+
+@pytest.mark.asyncio
+async def test_wildcard_with_duplicate_iteration_fix(coder):
+    """Test wildcard loading with duplicate iteration fix."""
+    server1 = MagicMock()
+    server1.name = "server1"
+    server1.config = {"enabled": True}
+    server2 = MagicMock()
+    server2.name = "server2"
+    server2.config = {"enabled": True}
+    coder.mcp_manager.servers = [server1, server2]
+    # server1 already loaded, server2 not loaded
+    coder.mcp_manager.connected_servers = {"server1": server1}
+    coder.mcp_manager.get_server.side_effect = lambda name: next(
+        (s for s in [server1, server2] if s.name == name), None
+    )
+    connect_calls = []
+
+    async def mock_connect_server(server_name):
+        connect_calls.append(server_name)
+        if server_name == "server2":
+            return True
+        return False
+
+    coder.mcp_manager.connect_server = mock_connect_server
+
+    # Mock interruptible to just execute the coroutine
+    async def mock_interruptible(coro, event):
+        return await coro, False
+
+    coder.coroutines.interruptible = mock_interruptible
+    result = await ResourceManagerTool.execute(coder, load_mcp=["*"])
+    # Should only attempt to load server2 (server1 should be skipped)
+    # server1 is already connected so it's skipped silently by wildcard expansion
+    assert "Server already loaded: server1" not in result
+    assert "Loaded server: server2" in result
+    assert connect_calls == ["server2"]  # Only server2 should have been connected
diff --git a/tests/unit/test_remove_mcp.py b/tests/unit/test_remove_mcp.py
new file mode 100644
index 00000000000..635fddcbbdc
--- /dev/null
+++ b/tests/unit/test_remove_mcp.py
@@ -0,0 +1,273 @@
+"""Unit tests for RemoveMcpTool.execute."""
+
+from unittest.mock import AsyncMock, MagicMock
+
+import pytest
+
+from cecli.tools.resource_manager import Tool as ResourceManagerTool
+
+
+class DummyIO:
+    """Mock IO object for testing."""
+
+    def __init__(self):
+        self.tool_error = MagicMock()
+        self.tool_warning = MagicMock()
+        self.tool_output = MagicMock()
+        self.interrupt_event = MagicMock()
+
+
+class DummyCoder:
+    """Mock Coder object for testing."""
+
+    def __init__(self):
+        self.io = DummyIO()
+        self.mcp_manager = MagicMock()
+        self.mcp_manager.servers = []
+        self.mcp_manager.connected_servers = {}
+        self.coroutines = MagicMock()
+        self.interrupt_event = MagicMock()
+
+
+@pytest.fixture
+def coder():
+    """Provide a dummy coder for testing."""
+    return DummyCoder()
+
+
+@pytest.fixture
+def mock_server():
+    """Provide a mock MCP server."""
+    server = MagicMock()
+    server.name = "test-server"
+    return server
+
+
+@pytest.mark.asyncio
+async def test_remove_mcp_tool_success():
+    """Test successful removal of an MCP server."""
+    # Setup
+    coder = MagicMock()
+    coder.agent_config = {"include_context_blocks": {"servers"}, "exclude_context_blocks": set()}
+    coder.mcp_manager = MagicMock()
+    server = MagicMock()
+    server.name = "test-server"
+    coder.mcp_manager.get_server.return_value = server
+    coder.mcp_manager.connected_servers = {"test-server": server}
+
+    # Mock disconnect_server as an async function that returns (True, False)
+    async def mock_disconnect(server_name):
+        return True, False
+
+    coder.mcp_manager.disconnect_server = AsyncMock(side_effect=mock_disconnect)
+
+    # Mock the interruptible method to execute the coroutine directly without interruption
+    async def mock_interruptible(coro, event):
+        return await coro
+
+    coder.coroutines = MagicMock()
+    coder.coroutines.interruptible = mock_interruptible
+    coder.interrupt_event = MagicMock()
+    # Execute
+    result = await ResourceManagerTool.execute(coder, remove_mcp=["test-server"])
+    # Assertions
+    assert "Removed server: test-server" in result
+    coder.mcp_manager.disconnect_server.assert_awaited_once_with("test-server")
+
+
+@pytest.mark.asyncio
+async def test_remove_mcp_tool_non_existent():
+    """Test removing a non-existent MCP server."""
+    # Setup
+    coder = MagicMock()
+    coder.agent_config = {"include_context_blocks": {"servers"}, "exclude_context_blocks": set()}
+    coder.mcp_manager = MagicMock()
+    # Create a mock server that exists (to bypass the 'no servers' check)
+    existing_server = MagicMock()
+    existing_server.name = "existing-server"
+    coder.mcp_manager.servers = [existing_server]
+    # But the one we're looking for doesn't exist
+    coder.mcp_manager.get_server.return_value = None
+    # Execute
+    result = await ResourceManagerTool.execute(coder, remove_mcp=["non-existent-server"])
+    # Assertions
+    assert "MCP server non-existent-server does not exist." in result
+
+
+@pytest.mark.asyncio
+async def test_remove_mcp_tool_not_connected():
+    """Test removing a server that is not connected."""
+    coder = MagicMock()
+    coder.agent_config = {"include_context_blocks": {"servers"}, "exclude_context_blocks": set()}
+    coder.mcp_manager = MagicMock()
+    server = MagicMock()
+    server.name = "test-server"
+    coder.mcp_manager.servers = [server]
+    coder.mcp_manager.get_server.return_value = server
+    coder.mcp_manager.connected_servers = {}
+    result = await ResourceManagerTool.execute(coder, remove_mcp=["test-server"])
+    assert "Server test-server is not currently connected." in result
+
+
+@pytest.mark.asyncio
+async def test_remove_mcp_tool_wildcard():
+    """Test removing all servers with wildcard '*'."""
+    coder = MagicMock()
+    coder.agent_config = {"include_context_blocks": {"servers"}, "exclude_context_blocks": set()}
+    coder.mcp_manager = MagicMock()
+    server1 = MagicMock()
+    server1.name = "server1"
+    server2 = MagicMock()
+    server2.name = "server2"
+    coder.mcp_manager.servers = [server1, server2]
+    coder.mcp_manager.connected_servers = {"server1": server1, "server2": server2}
+    coder.mcp_manager.get_server.side_effect = lambda name: next(
+        (s for s in [server1, server2] if s.name == name), None
+    )
+
+    # Mock disconnect_server as an async function that returns (True, False)
+    async def mock_disconnect(server_name):
+        return True, False
+
+    coder.mcp_manager.disconnect_server = AsyncMock(side_effect=mock_disconnect)
+
+    # Mock interruptible to execute the coroutine without interruption
+    async def mock_interruptible(coro, event):
+        return await coro
+
+    coder.coroutines = MagicMock()
+    coder.coroutines.interruptible = mock_interruptible
+    coder.interrupt_event = MagicMock()
+    result = await ResourceManagerTool.execute(coder, remove_mcp=["*"])
+    assert "Removed server: server1" in result
+    assert "Removed server: server2" in result
+
+
+@pytest.mark.asyncio
+async def test_remove_mcp_tool_interrupted():
+    """Test when removal is interrupted."""
+    coder = MagicMock()
+    coder.agent_config = {"include_context_blocks": {"servers"}, "exclude_context_blocks": set()}
+    coder.mcp_manager = MagicMock()
+    server = MagicMock()
+    server.name = "test-server"
+    coder.mcp_manager.servers = [server]
+    coder.mcp_manager.get_server.return_value = server
+    coder.mcp_manager.connected_servers = {"test-server": server}
+
+    async def mock_disconnect(server_name):
+        return False, True
+
+    coder.mcp_manager.disconnect_server = AsyncMock(side_effect=mock_disconnect)
+
+    async def mock_interruptible(coro, event):
+        return False, True
+
+    coder.coroutines.interruptible = mock_interruptible
+    coder.interrupt_event = MagicMock()
+    result = await ResourceManagerTool.execute(coder, remove_mcp=["test-server"])
+    assert "Interrupted: test-server" in result
+
+
+@pytest.mark.asyncio
+async def test_remove_mcp_tool_failed():
+    """Test when removal fails."""
+    coder = MagicMock()
+    coder.agent_config = {"include_context_blocks": {"servers"}, "exclude_context_blocks": set()}
+    coder.mcp_manager = MagicMock()
+    server = MagicMock()
+    server.name = "test-server"
+    coder.mcp_manager.servers = [server]
+    coder.mcp_manager.get_server.return_value = server
+    coder.mcp_manager.connected_servers = {"test-server": server}
+
+    async def mock_disconnect(server_name):
+        return False, False
+
+    coder.mcp_manager.disconnect_server = AsyncMock(side_effect=mock_disconnect)
+
+    async def mock_interruptible(coro, event):
+        return await coro
+
+    coder.coroutines.interruptible = mock_interruptible
+    coder.interrupt_event = MagicMock()
+    result = await ResourceManagerTool.execute(coder, remove_mcp=["test-server"])
+    assert "Unable to remove server: test-server" in result
+
+
+@pytest.mark.asyncio
+async def test_remove_mcp_tool_no_servers_configured():
+    """Test when no MCP servers are configured at all."""
+    coder = MagicMock()
+    coder.agent_config = {"include_context_blocks": {"servers"}, "exclude_context_blocks": set()}
+    coder.mcp_manager = MagicMock()
+    coder.mcp_manager.servers = []
+    result = await ResourceManagerTool.execute(coder, remove_mcp=["test"])
+    assert result == "No MCP servers are configured."
+
+
+@pytest.mark.asyncio
+async def test_remove_mcp_tool_mixed_results():
+    """Test mixed success/failure results."""
+    coder = MagicMock()
+    coder.agent_config = {"include_context_blocks": {"servers"}, "exclude_context_blocks": set()}
+    coder.mcp_manager = MagicMock()
+    server1 = MagicMock()
+    server1.name = "server1"
+    server2 = MagicMock()
+    server2.name = "server2"
+    coder.mcp_manager.servers = [server1, server2]
+    coder.mcp_manager.connected_servers = {"server1": server1, "server2": server2}
+    coder.mcp_manager.get_server.side_effect = lambda name: next(
+        (s for s in [server1, server2] if s.name == name), None
+    )
+    call_count = 0
+
+    async def mock_disconnect(server_name):
+        nonlocal call_count
+        result = (True, False) if call_count == 0 else (False, False)
+        call_count += 1
+        return result
+
+    coder.mcp_manager.disconnect_server = AsyncMock(side_effect=mock_disconnect)
+
+    async def mock_interruptible(coro, event):
+        return await coro
+
+    coder.coroutines.interruptible = mock_interruptible
+    coder.interrupt_event = MagicMock()
+    result = await ResourceManagerTool.execute(coder, remove_mcp=["server1", "server2"])
+    assert "Removed server: server1" in result
+    assert "Unable to remove server: server2" in result
+
+
+@pytest.mark.asyncio
+async def test_remove_mcp_tool_dictionary_iteration_fix():
+    """Test that dictionary iteration bug is fixed - iterates over keys correctly."""
+    coder = MagicMock()
+    coder.agent_config = {"include_context_blocks": {"servers"}, "exclude_context_blocks": set()}
+    coder.mcp_manager = MagicMock()
+    server1 = MagicMock()
+    server1.name = "server1"
+    server2 = MagicMock()
+    server2.name = "server2"
+    coder.mcp_manager.servers = [server1, server2]
+    coder.mcp_manager.connected_servers = {"server1": server1, "server2": server2}
+    coder.mcp_manager.get_server.side_effect = lambda name: next(
+        (s for s in [server1, server2] if s.name == name), None
+    )
+
+    async def mock_disconnect(server_name):
+        return True, False
+
+    coder.mcp_manager.disconnect_server = AsyncMock(side_effect=mock_disconnect)
+
+    async def mock_interruptible(coro, event):
+        return await coro
+
+    coder.coroutines.interruptible = mock_interruptible
+    coder.interrupt_event = MagicMock()
+    result = await ResourceManagerTool.execute(coder, remove_mcp=["*"])
+    # Should successfully remove both servers using dictionary keys
+    assert "Removed server: server1" in result
+    assert "Removed server: server2" in result
diff --git a/tests/unit/test_unit_load_mcp_tool.py b/tests/unit/test_unit_load_mcp_tool.py
new file mode 100644
index 00000000000..9458fe5bdf5
--- /dev/null
+++ b/tests/unit/test_unit_load_mcp_tool.py
@@ -0,0 +1,138 @@
+"""Unit tests for load-mcp tool."""
+
+from unittest.mock import AsyncMock, MagicMock
+
+import pytest
+
+from cecli.tools.resource_manager import Tool as ResourceManagerTool
+
+
+@pytest.fixture
+def mock_mcp_manager():
+    """Fixture for a mocked McpServerManager."""
+    manager = MagicMock()
+    manager.connected_servers = {}
+
+    # Mock servers
+    server1 = MagicMock()
+    server1.name = "test-server"
+    server1.config = {"enabled": True}
+
+    server2 = MagicMock()
+    server2.name = "server2"
+    server2.config = {"enabled": True}
+
+    server3 = MagicMock()
+    server3.name = "server3"
+    server3.config = {"enabled": False}
+
+    manager.servers = [server1, server2, server3]
+
+    def get_server_side_effect(name):
+        if name == "test-server":
+            return server1
+        if name == "server2":
+            return server2
+        if name == "server3":
+            return server3
+        return None
+
+    manager.get_server.side_effect = get_server_side_effect
+
+    async def connect(server_name):
+        manager.connected_servers[server_name] = "connected"
+        return True, False  # (did_connect, interrupted)
+
+    async def disconnect(server_name):
+        if server_name in manager.connected_servers:
+            del manager.connected_servers[server_name]
+            return True, False
+        return False, False
+
+    manager.connect_server = AsyncMock(side_effect=connect)
+    manager.disconnect_server = AsyncMock(side_effect=disconnect)
+    manager.add_server = AsyncMock()
+    return manager
+
+
+@pytest.mark.asyncio
+async def test_load_mcp_tool_success(mock_mcp_manager):
+    """Test loading a single MCP server successfully."""
+
+    # Mock the coder
+    coder = MagicMock()
+    coder.agent_config = {"include_context_blocks": {"servers"}, "exclude_context_blocks": set()}
+    coder.mcp_manager = mock_mcp_manager
+
+    # Mock interruptible to return (await coro, False)
+    async def mock_interruptible(coro, event):
+        return await coro
+
+    coder.coroutines = MagicMock()
+    coder.coroutines.interruptible.side_effect = mock_interruptible
+    coder.interrupt_event = MagicMock()
+    coder.registered_servers = {"included": set(), "excluded": set()}
+    result = await ResourceManagerTool.execute(coder, load_mcp=["test-server"])
+
+    assert "Loaded server: test-server" in result
+    mock_mcp_manager.connect_server.assert_awaited_once_with("test-server")
+
+
+@pytest.mark.asyncio
+async def test_load_mcp_tool_non_existent(mock_mcp_manager):
+    """Test loading a non-existent MCP server."""
+
+    coder = MagicMock()
+    coder.agent_config = {"include_context_blocks": {"servers"}, "exclude_context_blocks": set()}
+    coder.mcp_manager = mock_mcp_manager
+
+    result = await ResourceManagerTool.execute(coder, load_mcp=["non-existent-server"])
+
+    assert "MCP server non-existent-server does not exist." in result
+    mock_mcp_manager.connect_server.assert_not_awaited()
+
+
+@pytest.mark.asyncio
+async def test_load_mcp_tool_already_loaded(mock_mcp_manager):
+    """Test loading an already loaded MCP server."""
+    coder = MagicMock()
+    coder.agent_config = {"include_context_blocks": {"servers"}, "exclude_context_blocks": set()}
+    coder.mcp_manager = mock_mcp_manager
+    # Pre-populate connected_servers
+    server = mock_mcp_manager.get_server("test-server")
+    coder.mcp_manager.connected_servers = {"test-server": server}
+
+    result = await ResourceManagerTool.execute(coder, load_mcp=["test-server"])
+
+    assert "Server already loaded: test-server" in result
+    mock_mcp_manager.connect_server.assert_not_awaited()
+
+
+@pytest.mark.asyncio
+async def test_load_mcp_tool_wildcard_and_duplicate_fix(mock_mcp_manager):
+    """Test loading with wildcard and duplicate fix."""
+    coder = MagicMock()
+    coder.agent_config = {"include_context_blocks": {"servers"}, "exclude_context_blocks": set()}
+    coder.mcp_manager = mock_mcp_manager
+
+    # Mock interruptible to return (await coro, False)
+    async def mock_interruptible(coro, event):
+        return await coro
+
+    coder.coroutines = MagicMock()
+    coder.coroutines.interruptible.side_effect = mock_interruptible
+    coder.interrupt_event = MagicMock()
+
+    # Set up connected_servers: server1 is already connected
+    server1 = mock_mcp_manager.get_server("test-server")
+    coder.mcp_manager.connected_servers = {"test-server": server1}
+
+    result = await ResourceManagerTool.execute(coder, load_mcp=["*"])
+
+    # Check results
+    # Wildcard expansion skips already-connected servers; no "already loaded" message is produced
+    assert "Loaded server: server2" in result
+    # Non-enabled servers are filtered out silently by wildcard expansion
+
+    # Verify connect_server was called only once for server2
+    mock_mcp_manager.connect_server.assert_awaited_once_with("server2")
diff --git a/tests/unit/test_unit_remove_mcp_tool.py b/tests/unit/test_unit_remove_mcp_tool.py
new file mode 100644
index 00000000000..822841ecb86
--- /dev/null
+++ b/tests/unit/test_unit_remove_mcp_tool.py
@@ -0,0 +1,293 @@
+"""Unit tests for RemoveMcpTool.execute."""
+
+from unittest.mock import AsyncMock, MagicMock
+
+import pytest
+
+from cecli.tools.resource_manager import Tool as ResourceManagerTool
+
+
+class DummyIO:
+    """Mock IO object for testing."""
+
+    def __init__(self):
+        self.tool_error = MagicMock()
+        self.tool_warning = MagicMock()
+        self.tool_output = MagicMock()
+        self.interrupt_event = MagicMock()
+
+
+class DummyCoder:
+    """Mock Coder object for testing."""
+
+    def __init__(self):
+        self.io = DummyIO()
+        self.mcp_manager = MagicMock()
+        self.mcp_manager.servers = []
+        self.mcp_manager.connected_servers = {}
+        self.coroutines = MagicMock()
+        self.interrupt_event = MagicMock()
+        self.agent_config = {"include_context_blocks": {"servers"}, "exclude_context_blocks": set()}
+
+
+@pytest.fixture
+def coder():
+    """Provide a dummy coder for testing."""
+    return DummyCoder()
+
+
+@pytest.fixture
+def mock_server():
+    """Provide a mock MCP server."""
+    server = MagicMock()
+    server.name = "test-server"
+    return server
+
+
+@pytest.mark.asyncio
+async def test_remove_mcp_tool_success():
+    """Test successful removal of an MCP server."""
+    # Setup
+    coder = MagicMock()
+    coder.agent_config = {"include_context_blocks": {"servers"}, "exclude_context_blocks": set()}
+    coder.mcp_manager = MagicMock()
+    server = MagicMock()
+    server.name = "test-server"
+    coder.mcp_manager.get_server.return_value = server
+    coder.mcp_manager.connected_servers = {"test-server": server}
+
+    # Mock disconnect_server as an AsyncMock that returns (True, False)
+    async def mock_disconnect(server_name):
+        return True, False
+
+    coder.mcp_manager.disconnect_server = AsyncMock(side_effect=mock_disconnect)
+
+    # Mock the interruptible method to execute the coroutine directly without interruption
+    async def mock_interruptible(coro, event):
+        return await coro
+
+    coder.coroutines = MagicMock()
+    coder.coroutines.interruptible = mock_interruptible
+    coder.interrupt_event = MagicMock()
+
+    # Execute
+    result = await ResourceManagerTool.execute(coder, remove_mcp=["test-server"])
+
+    # Assertions
+    assert "Removed server: test-server" in result
+    coder.mcp_manager.disconnect_server.assert_awaited_once_with("test-server")
+
+
+@pytest.mark.asyncio
+async def test_remove_mcp_tool_non_existent():
+    """Test removing a non-existent MCP server."""
+    # Setup
+    coder = MagicMock()
+    coder.agent_config = {"include_context_blocks": {"servers"}, "exclude_context_blocks": set()}
+    coder.mcp_manager = MagicMock()
+    # Create a mock server that exists (to bypass the 'no servers' check)
+    existing_server = MagicMock()
+    existing_server.name = "existing-server"
+    coder.mcp_manager.servers = [existing_server]
+    # But the one we're looking for doesn't exist
+    coder.mcp_manager.get_server.return_value = None
+
+    # Execute
+    result = await ResourceManagerTool.execute(coder, remove_mcp=["non-existent-server"])
+
+    # Assertions
+    assert "MCP server non-existent-server does not exist." in result
+
+
+@pytest.mark.asyncio
+async def test_remove_mcp_tool_not_connected():
+    """Test removing a server that is not connected."""
+    coder = MagicMock()
+    coder.agent_config = {"include_context_blocks": {"servers"}, "exclude_context_blocks": set()}
+    coder.mcp_manager = MagicMock()
+    server = MagicMock()
+    server.name = "test-server"
+    coder.mcp_manager.servers = [server]
+    coder.mcp_manager.get_server.return_value = server
+    coder.mcp_manager.connected_servers = {}
+
+    result = await ResourceManagerTool.execute(coder, remove_mcp=["test-server"])
+
+    assert "Server test-server is not currently connected." in result
+
+
+@pytest.mark.asyncio
+async def test_remove_mcp_tool_wildcard():
+    """Test removing all servers with wildcard '*'."""
+    coder = MagicMock()
+    coder.agent_config = {"include_context_blocks": {"servers"}, "exclude_context_blocks": set()}
+    coder.mcp_manager = MagicMock()
+
+    server1 = MagicMock()
+    server1.name = "server1"
+    server2 = MagicMock()
+    server2.name = "server2"
+    coder.mcp_manager.servers = [server1, server2]
+    coder.mcp_manager.connected_servers = {"server1": server1, "server2": server2}
+    coder.mcp_manager.get_server.side_effect = lambda name: next(
+        (s for s in [server1, server2] if s.name == name), None
+    )
+
+    # Mock disconnect_server as an AsyncMock that returns (True, False)
+    async def mock_disconnect(server_name):
+        return True, False
+
+    coder.mcp_manager.disconnect_server = AsyncMock(side_effect=mock_disconnect)
+
+    async def mock_interruptible(coro, event):
+        return await coro
+
+    coder.coroutines = MagicMock()
+    coder.coroutines.interruptible = mock_interruptible
+    coder.interrupt_event = MagicMock()
+
+    result = await ResourceManagerTool.execute(coder, remove_mcp=["*"])
+
+    assert "Removed server: server1" in result
+    assert "Removed server: server2" in result
+
+
+@pytest.mark.asyncio
+async def test_remove_mcp_tool_interrupted():
+    """Test when removal is interrupted."""
+    coder = MagicMock()
+    coder.agent_config = {"include_context_blocks": {"servers"}, "exclude_context_blocks": set()}
+    coder.mcp_manager = MagicMock()
+    server = MagicMock()
+    server.name = "test-server"
+    coder.mcp_manager.servers = [server]
+    coder.mcp_manager.get_server.return_value = server
+    coder.mcp_manager.connected_servers = {"test-server": server}
+
+    async def mock_disconnect(server_name):
+        return False, True
+
+    coder.mcp_manager.disconnect_server = AsyncMock(side_effect=mock_disconnect)
+
+    async def mock_interruptible(coro, event):
+        return False, True
+
+    coder.coroutines.interruptible = mock_interruptible
+    coder.interrupt_event = MagicMock()
+
+    result = await ResourceManagerTool.execute(coder, remove_mcp=["test-server"])
+
+    assert "Interrupted: test-server" in result
+
+
+@pytest.mark.asyncio
+async def test_remove_mcp_tool_failed():
+    """Test when removal fails."""
+    coder = MagicMock()
+    coder.agent_config = {"include_context_blocks": {"servers"}, "exclude_context_blocks": set()}
+    coder.mcp_manager = MagicMock()
+    server = MagicMock()
+    server.name = "test-server"
+    coder.mcp_manager.servers = [server]
+    coder.mcp_manager.get_server.return_value = server
+    coder.mcp_manager.connected_servers = {"test-server": server}
+
+    async def mock_disconnect(server_name):
+        return False, False
+
+    coder.mcp_manager.disconnect_server = AsyncMock(side_effect=mock_disconnect)
+
+    async def mock_interruptible(coro, event):
+        return await coro
+
+    coder.coroutines.interruptible = mock_interruptible
+    coder.interrupt_event = MagicMock()
+
+    result = await ResourceManagerTool.execute(coder, remove_mcp=["test-server"])
+
+    assert "Unable to remove server: test-server" in result
+
+
+@pytest.mark.asyncio
+async def test_remove_mcp_tool_no_servers_configured():
+    """Test when no MCP servers are configured at all."""
+    coder = MagicMock()
+    coder.agent_config = {"include_context_blocks": {"servers"}, "exclude_context_blocks": set()}
+    coder.mcp_manager = MagicMock()
+    coder.mcp_manager.servers = []
+
+    result = await ResourceManagerTool.execute(coder, remove_mcp=["test"])
+
+    assert result == "No MCP servers are configured."
+
+
+@pytest.mark.asyncio
+async def test_remove_mcp_tool_mixed_results():
+    """Test mixed success/failure results."""
+    coder = MagicMock()
+    coder.agent_config = {"include_context_blocks": {"servers"}, "exclude_context_blocks": set()}
+    coder.mcp_manager = MagicMock()
+    server1 = MagicMock()
+    server1.name = "server1"
+    server2 = MagicMock()
+    server2.name = "server2"
+    coder.mcp_manager.servers = [server1, server2]
+    coder.mcp_manager.connected_servers = {"server1": server1, "server2": server2}
+    coder.mcp_manager.get_server.side_effect = lambda name: next(
+        (s for s in [server1, server2] if s.name == name), None
+    )
+
+    call_count = 0
+
+    async def mock_disconnect(server_name):
+        nonlocal call_count
+        result = (True, False) if call_count == 0 else (False, False)
+        call_count += 1
+        return result
+
+    coder.mcp_manager.disconnect_server = AsyncMock(side_effect=mock_disconnect)
+
+    async def mock_interruptible(coro, event):
+        return await coro
+
+    coder.coroutines.interruptible = mock_interruptible
+    coder.interrupt_event = MagicMock()
+
+    result = await ResourceManagerTool.execute(coder, remove_mcp=["server1", "server2"])
+
+    assert "Removed server: server1" in result
+    assert "Unable to remove server: server2" in result
+
+
+@pytest.mark.asyncio
+async def test_remove_mcp_tool_dictionary_iteration_fix():
+    """Test that dictionary iteration bug is fixed - iterates over keys correctly."""
+    coder = MagicMock()
+    coder.agent_config = {"include_context_blocks": {"servers"}, "exclude_context_blocks": set()}
+    coder.mcp_manager = MagicMock()
+    server1 = MagicMock()
+    server1.name = "server1"
+    server2 = MagicMock()
+    server2.name = "server2"
+    coder.mcp_manager.servers = [server1, server2]
+    coder.mcp_manager.connected_servers = {"server1": server1, "server2": server2}
+    coder.mcp_manager.get_server.side_effect = lambda name: next(
+        (s for s in [server1, server2] if s.name == name), None
+    )
+
+    async def mock_disconnect(server_name):
+        return True, False
+
+    coder.mcp_manager.disconnect_server = AsyncMock(side_effect=mock_disconnect)
+
+    async def mock_interruptible(coro, event):
+        return await coro, False
+
+    coder.coroutines.interruptible = mock_interruptible
+    coder.interrupt_event = MagicMock()
+
+    result = await ResourceManagerTool.execute(coder, remove_mcp=["*"])
+
+    # Should successfully remove both servers using dictionary keys
+    assert "Removed server: server1" in result
+    assert "Removed server: server2" in result