Does my code leave my machine?
Run locally and no — tiny-brain stays on your machine, and only your AI tool’s token calls leave, to a provider you’ve authorised. When you opt into a remote worker, that task’s code is pulled into a secure remote sandbox to run there, and nowhere else.Which AI tools does it work with?
Claude Code is first-class; Codex, Gemini CLI, and Cursor are supported. tiny-brain wraps the tool you already use — it doesn’t replace your LLM.Which models can I use?
Any. Claude, GPT, Gemini, or a local model — the loop is the quality bar, not the model, so you can swap or mix vendors without changing your standards.Is it really free?
Yes. tiny-brain is free, runs on your machine, and is bring-your-own-key — you use the AI tool and account you already have.How is this different from just using conventional commits?
Conventional commits are a naming convention; nothing checks that afeat:
was preceded by a failing test or followed by a review. tiny-brain keeps the
same commit vocabulary but makes it load-bearing: the hooks verify the
sequence, attach every commit to a planned task, and derive project state
from the history. The format is the same one your team already knows — it
just stops being decorative.
What should my team commit to git?
The plan markdown (docs/prd/, fix documents) and the repo context
(.tiny-brain/analysis.json, config.json, tech/) — that’s the shared
picture. Live projections are per-clone under .git/tiny-brain/ on current
versions and stay out of the tree automatically. If your repo was
initialised on an older version and has committed progress files, they’re
harmless to keep or to gitignore — see
the tracking model.
Can I use it on a docs or writing repo?
Yes —tiny-brain init --type prose skips the code analysis and test
expectations while keeping plans, tracking, and the loop.
How do I uninstall it?
Cleanly. It’s markdown and config in your repo plus~/.tiny-brain/ —
remove the plugin (tiny-brain mcp uninstall), delete those, and your repo
is exactly as it was.