Local-first AI review for serious codebases

Your coding agent should not grade its own homework.

Argus gives you an independent reviewer with structural maps, semantic search, git history, and diff-aware analysis, so the model reviewing the code is not the same one that just wrote it.

local-first OpenAI / Anthropic / Gemini one binary CLI + MCP independent review

Why Argus

AI code review that actually understands your codebase.

Most AI review workflows reuse the same shallow context that produced the patch. Argus takes the opposite stance: review should be independent, repository-aware, and composable enough to fit a real developer workflow.

Independent by design

One model writes; a different reviewer critiques. Mistakes don't get graded by their author.

Full-repo context

Structural maps, semantic search, git history, and cross-file analysis keep the review anchored in the actual codebase instead of just the diff.

Provider choice

Switch between OpenAI, Anthropic, and Gemini without changing the workflow. Gemini's free tier lets you try it without a paid API plan, but Google may use free-tier prompts and responses to improve its products and human reviewers may read them, so use a paid key for private code.

One binary, eleven subcommands

Map, diff, search, history, review, describe, feedback, hook, init, doctor, and mcp, an MCP server, can be used together or as Unix-style pieces.

Review surface

Move from raw diff review to evidence-backed review.

Argus is strongest when teams want more than a pass or fail. It gives the reviewer evidence about file relationships, history, and likely impact, so review comments can be specific instead of generic style feedback.

What Argus adds

  • Repository maps and code navigation context
  • Semantic search to find similar patterns before commenting
  • Git history signals for churn and ownership clues
  • Provider-agnostic review orchestration

Where it fits

Use it locally before a pull request, in CI as a review gate, or through MCP when you want an IDE or agent workflow to call into the same review engine.

Install

Start with the package manager you already trust.

Argus is published across Rust and npm surfaces. The fast path is to initialize config once and then pipe a diff into review.

npm

npx argus-ai init
export GEMINI_API_KEY=your-key
git diff HEAD~1 | npx argus-ai review --repo .

cargo

cargo install argus-ai
argus init
git diff HEAD~1 | argus review --repo .