Local-first AI code review
AI code review that runs on a local model and never uploads your source.
- Kind
- Open source developer tool
- Role
- Author and maintainer
- Status
- Published on npm
- Years
- 2026 — present
Context
The useful moment for a review comment is before the code is pushed, not when a reviewer reaches the pull request the following day.
The problem
Hosted AI review tools fix the timing and introduce a worse problem: they send proprietary diffs to somebody else's API. Most employment agreements and most client contracts forbid exactly that, so those tools go uninstalled in the environments where they would help most.
What I did
A git pre-push hook that reads the diff for the outgoing commits and reviews it before the push completes. Only the outgoing commit range is read; the repository as a whole is never sent anywhere.
Inference runs against a local model through the Ollama HTTP API on localhost. No outbound calls at any point, so the tool works with the machine offline — which is the only reason it can be used on work under NDA.
Distributed as an npm CLI, so installing it is one command in tooling developers already have configured, and the hook is version-controlled across a team rather than a script each person copies by hand.
It fails open on every error path. A review tool that can block a push gets uninstalled the first time it misfires during an incident, so advisory output is the only version that survives contact with a real team.
Where it landed
Published on npm in 2026, and I maintain it. Output quality is bounded by whatever hardware the developer happens to have. That is the honest price of the privacy guarantee, and it is also the reason the tool is permitted to run in the places a hosted reviewer is a non-starter.
Built with
- Node.js
- TypeScript
- Ollama
- Local LLMs
- Git hooks
- npm
Working on something in this shape? Tell me about it.
Book a call