Computer scientist (Diplom-Informatiker), 20+ years building software — most of it full-stack web. My bet for the AI-assisted era: the durable value isn't the code, it's the structured knowledge we hand the agents. So I'm building the layer that holds it.

A schema-agnostic graph engine. Agents write typed, validated entities into a knowledge graph stored as plain markdown in git — human-readable, diffable, and owned by you. No database to run, no vendor lock-in. It's MCP-native, so any agent can query and mutate the graph directly instead of scraping a transcript.

Where AI adoption in real teams tends to break is getting specification and domain knowledge into the agent reliably. That's the gap Memstead targets.

• Built in Rust — engine, CLI, and MCP server
• Markdown + git as the substrate — your data, your files
• Typed, validated relationships instead of extracted vector soup
• LLM-first by design, with a human oversight layer on top

→ github.com/memstead/memstead

Rust · AI agents · Model Context Protocol (MCP) · knowledge graphs · Claude Code · full-stack web

• Memstead: github.com/memstead/memstead