Your AI Wrote the Code. Now Make Different AI Review It.

What it does Chorus acts as a local peer-review panel for AI-written code. It sends a given diff or task to two-to-four different LLM reviewers—Claude, GPT, Gemini, or others—through the CLI tools you already own, and surfaces a verdict only when they agree. The whole thing runs on your laptop and counts against the subscription quotas you already pay for, so a typical review costs nothing out of pocket.

The interesting bit Instead of burning pay-per-token API keys, Chorus drives the headless CLIs of your existing AI subscriptions to generate reviews. By forcing different vendors to cross-examine each other, it turns your existing monthly quotas into a multi-model safety net that costs nothing extra per review.

Key highlights

• Local-first: your code stays on your laptop; Chorus routes to the AI CLIs you already trust rather than a new vendor cloud.
• Template-driven reviews: built-in YAML patterns like code-review, bug-diagnose, and red-green define reviewer count, vendor mix, and consensus rules.
• Persona “hats”: assign reviewers roles like Sentinel (security), Cartographer (cross-platform), or Accountant (cost regressions) to target specific blind spots.
• MCP-native: plugs into Claude Code, Codex, Cursor, and others as an MCP server, letting you summon a multi-AI panel without leaving your editor.
• Subscription-first economics: leans on existing Claude Pro / ChatGPT Plus / Gemini Advanced quotas so typical reviews cost $0 out of pocket; API-key fallback averages $0.30–$1.50 per run.

Caveats

• Codex CLI headless mode currently blocks MCP tool calls under any approval policy; scripted or CI usage requires a dangerous bypass flag until a fix lands (see issue #16).

Verdict Worth a look if you already use AI coding assistants and want a structured way to catch their blind spots before merging. Skip it if you don’t generate code with AI, or if you’re hoping for a fully managed cloud service—this is strictly local tooling.