timwuhaotian/the-pair

The Pair – Automated AI Pair‑Programming

What it is : The Pair is an open‑source desktop application (built with Tauri, Rust and React + TypeScript) that runs two AI coding agents side‑by‑side. One agent – the Mentor – is read‑only and plans or reviews code. The other – the Executor – actually writes code and runs commands. The two agents cross‑check each other’s output to catch the “hallucinations” that single‑agent tools can miss.

Why it matters : Most AI coding assistants (e.g., GitHub Copilot, Cursor) are a single model that both writes and reviews its own work, which can let mistakes slip through. By splitting the roles and letting two independent models validate each other, The Pair aims to reduce those errors while still letting developers stay hands‑off for long‑running tasks.

Key features

• Dual‑agent architecture – Mentor (plan + review) + Executor (write + run).
• Multi‑provider support – works with ten different provider CLIs (Claude Code, Codex, Gemini/Antigravity, Kimi, Pi, Kiro, Aider, Grok Build, Muse Code, plus any local model via Ollama).
• Local orchestration – the app runs on macOS, Windows and Linux; only the model API calls need internet.
• Session snapshot & recovery – full conversation and file‑change history can be restored after a crash or restart.
• Git integration – tracks file changes, shows inline diffs, and keeps a per‑pair runtime folder.
• Resource monitoring – per‑agent CPU/memory usage, stall detection, and token‑usage counters.
• Human‑in‑the‑loop controls – pause, reassign, or resume any pair; iteration budget defaults to 20 turns.
• Rich UI – unified console, activity view, command palette, sound cues, dark/light themes, and i18n (English, Chinese, Japanese, Korean).

How to get started

1. Install at least one supported AI provider CLI (e.g., npm install -g opencode-ai or follow the provider’s own installer).
2. Download the desktop app from the GitHub releases page (or build from source with npm run build:*).
3. Launch the app, create a new pair by selecting a project folder, describing the task, and picking a Mentor model and an Executor model.
4. Watch the agents collaborate; intervene via the UI or let them run autonomously.

Who it’s for

• Developers who want a more rigorous, automated code‑generation workflow.
• Teams that need to audit AI‑generated code before it lands in a repo.
• Anyone curious about multi‑agent AI collaboration and wants a visual, open‑source tool to experiment with.

License : Apache 2.0 (free and permissive).

---

All details are taken directly from the repository’s README; no additional claims are made.