Local-first desktop app for orchestrating multi-provider AI code reviewers (Claude, Codex, Gemini) with live robot characters, meeting synthesis, and follow-up chat

Agent Review Room

Local-first desktop app for orchestrating multiple LLM reviewers against a local repository. Each reviewer works independently, then a manager consolidates their findings into a final synthesis.

Prerequisites

• Node.js 18–22 (LTS recommended). Node 23+ is untested. Check with node --version.
• npm 9+ (ships with Node 18+)
• Git installed and in PATH
• At least one supported AI CLI installed and authenticated:
  • Gemini CLI (gemini) — npm install -g @anthropic-ai/claude-code
  • Codex CLI (codex) — npm install -g @openai/codex
  • Gemini CLI (gemini) — npm install -g @anthropic-ai/gemini-cli
• macOS: Xcode Command Line Tools (xcode-select --install) for Electron native deps
• Linux/WSL: libnss3 libatk-bridge2.0-0 libgtk-3-0 libgbm1 libasound2 (or libasound2t64 on Ubuntu 24.04)

Quick Start

rm -rf node_modules package-lock.json   # clean slate if switching platforms
npm install
npm run dev

No API keys are configured in the app -- each CLI manages its own authentication. The app itself does not run a hosted backend or built-in telemetry service, but review runs are executed through those local third-party CLIs, which may send prompts and repository context to their providers depending on your CLI/auth setup.

Demo

Setting The Run

Resume Talk With Manager

How It Works

1. Setup -- Pick a local git repo, choose a review target (working tree or ref range), write optional instructions, toggle PR format, configure 1-5 reviewer agents. Each agent is selected from a dropdown populated by bundled .md skill files (shipped with the app) or ~/.config/agent-review-room/skills/. Built-in agents: security, architecture, regr