Introduction
Local-first AI code review for your working changes, in the browser or the terminal.
Local AI code review
DiffgazerInstall the published CLI, then get a structured review from dedicated AI agents.
Correctness
Security
Performance
Simplicity
Tests
Quick start
Diffgazer is a local binary that reviews your uncommitted changes with AI. Run it inside any repository and it reads what you have changed, then hands back a structured review. Web mode is the default and opens in your browser; the terminal UI is opt-in and still in beta.
How review works
Diffgazer takes the diff of your working changes, gathers the surrounding code as context, and sends both to five review agents. Each looks at the change through one lens: correctness, security, performance, simplicity, and tests. Their issues come back as one report you read inline against the diff. For the full pipeline, see how it works.
Two ways to review
By default diffgazer starts a local web environment and opens it in your browser. That is the path most people should use. The same review also runs in a keyboard-first terminal UI behind the --tui flag, but that mode is an incomplete beta and not recommended yet. See web mode and the terminal UI for the details of each.
Local and private
Diffgazer runs on your machine. There is no remote Diffgazer service to deploy, and nothing about your diff is uploaded to a dashboard you do not control. Credentials are stored locally or read from your environment; when you start a review, the diff and prompt content go to the model provider you chose under that provider's policy. See privacy for what is sent and what is not.
Where to go next
- Installation: requirements and the global install
- Your first review: run Diffgazer and read the result
- How it works: the diff-to-report pipeline in detail
- Privacy: what stays local and what is sent
- CLI reference: every command-line option