szarkans/multi
24 stars · Last commit 2026-11-03
Use [Multi]ple AI agents for code-review, verifying and questions
README preview
<h1 align="center">multi</h1> <p align="center"><a href="README.ru.md">[🇷🇺 →]</a> · <a href="README.zh.md">[🇨🇳 →]</a></p> <p align="center">one code review, several AI models. including free ones.</p> --- one model misses things. another model may notice them. `multi` puts independent reviewers on the same code, checks what they found, and gives you one report — including the disagreements. **it found 29 real bugs where the built-in Claude Code review found 14.** in our 16-case evaluation, with similar Claude usage and roughly twice the wall time. [see the results and how we measured them](evals/RESULTS.md). you don't need a team of expensive models. add free reviewers from OpenCode or OpenRouter to the models you already use. even a free opinion is useful when it catches something the others missed. > [!TIP] > for simplicity reason ill use claude code type command (e.g. `/multi:setup`) but you can use whatever your harness supports - `$multi` in Codex, `@multi` in opencode and etc <h2 align="center">install & run</h2> **lazy ass**