Does Agent Brain send my code or memories to a cloud service?

No. Agent Brain is self-hosted. Storage, vector search, cache, and local model calls run on your machine or infrastructure.

Which agents can use it?

Claude Code, Codex, Cursor, Gemini, Aider, and custom scripts can connect through MCP or REST.

Troubleshooting

Stack will not start

Make sure Docker Desktop is running and has enough memory available. Agent Brain needs multiple local services plus Ollama models, so 8 GB RAM is recommended.

MCP tools do not appear

Confirm the URL is exactly:

http://localhost:3100/mcp

Restart the client after editing its MCP configuration. For Claude Code, run /mcp after restart to refresh the tool list.

REST calls fail

Check the brain-engine health endpoint:

curl http://localhost:9090/health

If it fails, inspect the Docker service logs and confirm the stack is running.

First recall is slow

The first embedding request may warm up Ollama. Subsequent calls should be faster after the model is loaded.

Ollama models are missing

Run the model pull script from the Agent Brain product repo:

./scripts/pull-models.sh

Cache misses too often

Check the configured similarity threshold. A very high threshold improves precision but reduces cache hits. Start from the documented default and tune with the Cache Monitor.