NanoClaw
Personal AI assistant running Claude in containers with isolated memory per context.
NanoClaw is the agent backend — the brain that Kleos and Nyx talk to.
A single Node.js orchestrator routes messages from WhatsApp, Telegram, Slack, Discord, and Gmail to isolated Claude agents running inside Docker containers. Each agent gets its own filesystem, its own memory, and its own CLAUDE.md defining who it is and what it knows. The Obsidian knowledge base mounts read-write into the main agent’s container, so every conversation draws on the same structured vault.
The architecture is intentionally simple: one message loop, one container per invocation, one queue per agent. Channels self-register at startup. Groups define agent identity. The orchestrator doesn’t know or care which channel a message came from — it routes to the right agent and lets the container do the work.
Stack: Node.js, Claude Agent SDK, Docker, SQLite, Obsidian (PARA vault)
Status: Running in production as my daily assistant. Five channels active. The main agent (Lyra) handles knowledge management, project tracking, content drafting, and dev session orchestration.
What’s next: A web API endpoint (POST /api/chat with SSE streaming) to serve Kleos and Nyx as first-party clients — same brain, richer surfaces.