The Pair
Two AI agents cross-check each other's code.
The Pair is a free, open-source AI pair programming desktop app for macOS, Windows and Linux that runs two AI coding agents on one task. A read-only Mentor plans and reviews while an Executor writes code and runs commands, so they cross-check each other's work and catch AI mistakes before they reach your codebase. It connects straight to any model endpoint — Claude, GPT, Gemini, Kimi, GLM, Qwen, or local models via Ollama — with no agent CLI to install.
Features
Dual-agent loop — a read-only Mentor plans and reviews, an Executor writes code and runs commands, and each checks the other's work.
No agent CLI to install — the app talks to model endpoints over HTTPS itself (OpenAI Chat, Anthropic, Responses and Command Code protocols) and runs the agents' tools inside the pair's workspace.
Bring your own endpoint — import the configuration of opencode, codex, claude, kimi, pi, omp, gemini or killi, add a key by hand, or sign in with a ChatGPT, Command Code, OpenCode Go or Muse Code subscription.
Any model in either role — Claude, GPT, Gemini, Grok, GLM, Qwen, Kimi, MiniMax, DeepSeek, or local models via Ollama — each with its own reasoning effort.
Acceptance checks — the Mentor runs tests and lint, then returns a verdict with risk, evidence and a next step.
One session console with live tool calls, per-turn token usage and estimated cost.
Git change tracking with inline diffs, a worktree per task, and session recovery after a restart or crash.
An optional plan gate and iteration cap that pause the run for your review.
English, 简体中文, 日本語, 한국어.
Key questions
What is The Pair?
The Pair is a free, open-source AI pair programming desktop app for macOS, Windows and Linux that runs two AI coding agents on one task. A read-only Mentor plans and reviews while an Executor writes code and runs commands, so they cross-check each other's work and catch AI mistakes before they reach your codebase. It connects straight to any model endpoint — Claude, GPT, Gemini, Kimi, GLM, Qwen, or local models via Ollama — with no agent CLI to install. Core workflow: Dual-agent loop — a read-only Mentor plans and reviews, an Executor writes code and runs commands, and each checks the other's work.
Which platforms does The Pair support?
The Pair supports macOS, Windows and Linux.
How does The Pair handle data?
The Pair stores all session data, conversation history, and configuration locally on your device. There is no account system, no cloud backend, and no server that receives your data. API keys for AI providers are stored in your local configuration and used only to communicate directly with the provider endpoints you specify. The Pair does not collect, transmit, or sell any user data. Your code, conversations, and agent sessions never leave your device except when sent directly to the AI provider endpoints you have configured (e.g., Anthropic, OpenAI, Google). The Pair has no analytics SDK, no telemetry, no crash reporting, and no tracking of any kind.
How is The Pair different from Cursor, GitHub Copilot, or Aider?
Cursor, GitHub Copilot, and Aider drive a single assistant that writes code and reviews its own output. The Pair separates the two jobs: the Executor changes code and runs commands, and a read-only Mentor, which can be a different model, plans the work and reviews every result before it is accepted.
Which AI models does The Pair support?
Any model reachable through an endpoint you connect, including GPT, Claude, Gemini, Grok, GLM, Qwen, Kimi, MiniMax and DeepSeek, plus local models via Ollama. Endpoints can be imported from the configuration of opencode, codex, claude, kimi, pi, omp, gemini or killi, added by hand with a base URL and API key, or signed in with a ChatGPT, Command Code, OpenCode Go or Muse Code subscription. The Mentor and the Executor can use different models.
Do I need Claude Code, Codex or another agent CLI installed?
No. Since version 3.0 The Pair talks to model endpoints directly over HTTPS and runs the agents' tools itself, scoped to the pair's workspace. It needs one model endpoint with a credential and nothing else.
Is The Pair free?
Yes. The Pair is open source under the Apache 2.0 license and free to download. You pay only for your own model usage, or nothing at all when you run local models with Ollama.
How does The Pair reduce AI coding hallucinations?
By splitting the work into two roles. The Executor writes and verifies code, while the read-only Mentor reads the diff, runs acceptance checks such as tests and lint, and returns a verdict with risk, evidence and a next step. Work that fails the review goes back for another iteration instead of being accepted.