nanoclaw-extras — plain-language guide

What is this?

NanoClaw is a house AI-agent tool that runs agents inside containers. This little repo adds one capability to it: letting those containerized agents ask the free, local AI models running on your own PC (via a program called Ollama) to do quick, cheap work — instead of always calling out to a paid cloud AI for every small task.

Think of it like giving your NanoClaw agent a second, cheaper coworker sitting right next to it that it can ask "hey, can you quickly summarize this?" instead of picking up the phone to a much more expensive consultant every time.

The two pieces

1. ollama-mcp.mjs — a small translator program. NanoClaw's agents speak a protocol called MCP; Ollama speaks plain web requests. This file sits in between and translates.
2. wire-ollama.sh — a one-time setup script. You run it once after setting up NanoClaw, and it tells NanoClaw "hey, here's a new tool your agents can use" and copies the translator program into place.

Do I need to touch this?

Only if you're running NanoClaw yourself and want its agents to be able to use your local Ollama models. If that's not you, there's nothing to do here — this repo doesn't run on its own; it's an add-on for another project.

Is anything sensitive in here?

No. There are no passwords, API keys, or tokens anywhere in this repo — the local AI model server it talks to doesn't require any. The only settings are "where is Ollama running" and "which model to use by default," both of which are safe to see in plain text.

Where to look next

• README.md — the technical version of this same overview, plus exact usage commands.
• docs/technical/ARCHITECTURE.md — diagrams of how the pieces talk to each other.
• TESTING.md — how we know this actually works (with the honest caveats).