When you hand a task to an AI coding agent, the change that comes back can run to hundreds or even thousands of lines. Terminal diffs make it hard to point at exactly what you want changed, so you end up re-explaining in prose: “the error handling here is sloppy.” Crit is an open-source CLI tool that replaces that back-and-forth with a review screen in your browser. Open the plan, the diff, or the running app your agent produced, click the line that bothers you, write a comment ─ and the agent reads it and starts fixing. It runs as a single binary, and local use requires no login and no daemon. Built by Tomasz Tomczyk and released under the MIT license.
Key Features
- Four modes for four kinds of review: Plans and docs (Markdown rendered with per-line comments), Code (syntax-highlighted branch diffs with Git, jj, and Sapling support), Live (proxies a running localhost app so you can comment on DOM elements), and Preview (renders generated static HTML in an iframe)
- Inline comments on lines and ranges: Click a line number, or drag to select a range, and attach a comment. The display resembles a GitHub pull request, and unresolved threads stay flagged until you resolve them
- Round-to-round diffs: Each time the agent returns a revision, you can see what changed since the previous review as a split or unified diff, so you can visually confirm your feedback landed
- Support for 13+ agents: First-class plugins exist for Claude Code, Cursor, GitHub Copilot, Codex, OpenCode, Aider, Cline, Windsurf, Gemini, Qwen, and more. Any agent that can read a file and run a shell command can also be wired up
- Two-way GitHub / GitLab sync: Review comments can be synced with pull requests and merge requests, so the tool does not cut you off from your team’s existing flow
- Local-first design: The server binds to 127.0.0.1 and collects no telemetry. Sharing happens only when you explicitly upload
- Sharing for async review, plus Vim keybindings: Upload a review to crit.md to show it to others, and navigate with keyboard shortcuts such as
j/kto move andcto comment
Pricing
| Plan | Price | What you get |
|---|---|---|
| Local use | Free (MIT license) | All review modes, agent integrations, GitHub / GitLab sync. No account required |
| Sharing on crit.md | Free account | Upload a review to crit.md to share it. Sign in with GitHub OAuth |
| Self-hosting | Free (your own infrastructure costs only) | Deploy crit-web via Docker and point sharing at your own server |
Pricing is current as of August 2026. Check the official site for the latest information.
Pros & Cons
✅ Pros
- Large changes that are hard to follow in a terminal diff can be read in a familiar browser UI
- “This line, right here” goes straight to the agent, so you spend less effort rephrasing feedback in prose
- Installation is a single binary, available via Homebrew, Go, Nix, or a Windows download
- Fully local with no telemetry, so internal code never leaves your machine
- MIT-licensed open source, and even the sharing backend can be self-hosted
⚠️ Cons
- It is a CLI tool, so people uncomfortable with the terminal face a setup hurdle
- Each supported agent has its own plugin installation steps, which adds friction the first time
- “Send to Agent,” which lets the agent respond to comments on the spot, is labeled experimental
- Releases come frequently, so features and settings can shift between versions
- Using the sharing feature means uploading to crit.md, or standing up a self-hosted instance
Comparison with Similar Services
| Criteria | Crit | GitHub pull requests | CodeRabbit | Built-in agent diff views |
|---|---|---|---|---|
| Main purpose | Review loop with an agent | Human-to-human code review | Automated AI review | Checking what changed |
| Who you review | An AI agent | A colleague | A human (AI flags issues) | An AI agent |
| Where comments go | A structured file the agent reads | Discussion on the pull request | Comments on the pull request | Usually the chat pane |
| Usable before committing | Yes (local working tree) | Generally after pushing | Generally after opening a PR | Yes |
| Runtime | Local single binary | Cloud | Cloud | Inside the editor / terminal |
| Cost | Free (MIT) | Follows your GitHub plan | Free tier and paid plans | Follows each tool |
Who Is It For
- Developers handing sizable tasks to Claude Code or Cursor who now spend real time reviewing the diffs that come back
- People who want to read the agent’s plan before implementation and correct the approach in detail
- Front-end developers who want to look at the UI and say “tighten this spacing” against a specific element
- Teams that want a review environment without sending internal code to an external service
- Anyone who wants to keep their existing GitHub / GitLab flow and simply add a review path aimed at agents
Summary
Crit is a tool for reading, pointing at, and fixing the large volume of change an AI agent produces. Its core value is putting a UI modeled on human code review on your own machine and turning line-level remarks directly into instructions for the agent. It covers a wide range of targets ─ plans, code, running apps, and static HTML ─ and the combination of no telemetry, an MIT license, and self-hosting makes it easy to justify at work. If the amount you delegate to agents has grown to the point where review can no longer keep up, brew install crit is a low-cost place to start.