feat: agentic server Phase 1 — context.agent injection + standalone LLM gateway

[pyramation]

Summary

Adds context.agent to FunctionContext — every cloud function (Node.js and Python) now has a fully typed, tenant-isolated LLM inference client injected automatically. Also creates @constructive-io/agentic-server as a standalone OpenAI-compatible proxy that spins up with docker compose up on port 3003.

Three-layer architecture:

1. Typed context — context.agent.inference({ messages, model }) / context.agent.embed(input) in fn-runtime
2. Standalone service — packages/agentic-server/ proxies to any LLM provider (OpenAI, Ollama, Anthropic)
3. Header security — X-Database-Id/X-Entity-Id/X-Actor-Id set server-side from job context (unforgeable); external requests have identity headers stripped

Key interfaces:

// fn-types: AgentContext added to FunctionContext
type AgentContext = {
  inference(opts: InferenceOptions): Promise<InferenceResult>;
  embed(input: string | string[], model?: string): Promise<EmbedResult>;
  readonly databaseId: string | undefined;
  readonly entityId: string | undefined;
};

// Usage in any handler:
const handler: FunctionHandler = async (params, context) => {
  const result = await context.agent.inference({
    messages: [{ role: 'user', content: 'Hello' }],
    model: 'gpt-4o'
  });
  return { reply: result.content };
};

fn-runtime wiring: buildContext() calls createAgentContext(env.AGENTIC_SERVER_URL, { databaseId, entityId, actorId }). When AGENTIC_SERVER_URL is unset, methods throw with a descriptive error; databaseId/entityId are still readable.

Agentic server: OpenAI-compatible proxy (/v1/chat/completions, /v1/embeddings, /healthz) with per-provider request/response transformation (Ollama ↔ OpenAI format normalization). Header middleware strips X-Database-Id from external requests (only trusts X-Internal-Service: fn-runtime).

Docker Compose: agentic-server service added (port 3003), LLM_PROVIDER_URL/LLM_PROVIDER_TYPE/LLM_PROVIDER_API_KEY env vars. make dev-fn auto-sets AGENTIC_SERVER_URL=http://localhost:3003.

Python template: templates/python/agent.py — AgentContext dataclass with from_request(request), inference(), embed(). Injected into handler as agent kwarg.

Tests: 7 new integration tests (TDD) covering inference routing, embed routing, header forwarding, stub behavior, and security.

E2E fix: Added connection retry logic (getTestConnections() retries 10× with 3s intervals) and 90s beforeAll timeout to all E2E tests. This fixes a pre-existing flaky timeout when the K8s PostgreSQL pod isn't ready within 30s (observed on main run 26556238326 and other branches).

Refs: constructive-planning#1060

Link to Devin session: https://app.devin.ai/sessions/b2291a8e333e445aa125a2efd1996206
Requested by: @pyramation

[devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment, CI, and merge conflict monitoring

[socket-security]

Review the following changes in direct dependencies. Learn more about Socket for GitHub.

Diff	Package	Supply Chain Security	Vulnerability	Quality	Maintenance	License
	npm/​ts-node@​10.9.2

View full report

[devin-ai-integration]

Superseded — already merged into develop via #104