What is veto-gate?

veto-gate is a review gate that sits in front of git commit when you are working inside Claude Code. Before a commit is allowed through, the staged changes are sent to an AI reviewer, and if the reviewer spots a real problem, the commit is blocked with an explanation instead of quietly landing. The idea is to catch broken or invented code before it gets into your history, not after. The main reviewer is the OpenAI Codex CLI, which you connect using your own OpenAI account since veto-gate does not pay for any AI calls on your behalf. In front of that main reviewer you can optionally add a free pre-checker, either a local model running on your own machine through a tool like LM Studio or Ollama, or a remote free-tier option through Groq or Gemini. The pre-checker looks at the diff first and catches obvious mistakes early, saving the main reviewer's usage for trickier cases. A few checks run without any AI at all: a size limit on how large a single commit can be, a check for invented function calls, and checks around tests and evidence notes. Setup on macOS involves installing a few common command line tools, installing and logging into the Codex CLI, cloning the repository, and running an install script. After that, each repository you want reviewed needs a small config file turned on, since the gate only watches repositories that opt in. There is also a local web dashboard you can start that shows recent runs and any findings, along with per-repository settings. This is a shell-based developer tool aimed at people already using Claude Code and the Codex CLI who want an automatic safety check before commits land. It has been developed and tested on macOS, with Windows support available only through WSL2 and not yet fully verified. The project is released under the MIT license.