Your terminal.
Any model.
One agent.
Copair is an open source AI coding agent that works with Claude, GPT-5, Gemini, Llama 4, DeepSeek, or any OpenAI-compatible model. No vendor lock-in.
See It In Action
Watch Copair work through real-world coding tasks
Debugging & Refactoring
See how Copair analyzes code, identifies issues, and proposes refactoring suggestions in real-time.
Building Features
Watch Copair write tests, implement features, and integrate them into your existing codebase automatically.
Git Integration
Copair creates branches, commits changes with meaningful messages, and manages your git workflow seamlessly.
Model Switching
Dynamically switch between Claude, GPT, Gemini, and other models within the same session using /model command.
Key Features
Powerful capabilities built for modern development workflows
Any Model, One Interface
Works with Anthropic Claude, OpenAI GPT-5, Google Gemini, and any OpenAI-compatible endpoint (Ollama, vLLM, LM Studio, llama.cpp). Switch models mid-session with /model.
Built-in Developer Tools
8 tools out of the box: Read, Write, Edit, Grep, Glob, Bash, Git, WebSearch. The agent reads your codebase, writes code, runs tests, searches the web, and commits changes.
Rich Terminal UI
Ink-powered React UI with bordered input box, syntax-highlighted code blocks, inline file diffs, real-time streaming, context window percentage, and vi-mode editing.
Session Persistence
Sessions survive exits and crashes. Resume where you left off with copair --resume. Sessions are auto-named from your git branch and summarized on exit.
Extensible via MCP
Connect any Model Context Protocol server to add custom tools. Tools are discovered at startup and namespaced automatically.
Secure by Default
Three permission modes: ask (default), auto-approve, deny. Tool calls require explicit approval. No telemetry, no analytics. Your code stays on your machine.
Built with security in mind
Copair is a local agent. Your code, keys, and session data stay on your machine.
Project boundary enforcement
The agent stays inside your repo. Reading or writing outside the project root requires explicit config.
Approval gate
Every write, edit, and shell command is shown to you before it runs. Nothing executes without your say.
Secrets never logged
API keys and tokens are redacted from all logs, session files, and model output before they are stored.
Open source
Full source on GitHub. No hidden telemetry, no analytics, no callbacks home. Audit it yourself.
Supported Providers
Works with your favorite LLM ecosystem
| Provider | Models | Type | Tool Calling | Streaming |
|---|---|---|---|---|
Claude Opus 4.6Claude Sonnet 4.6Claude Haiku 4.5 | Native | ✓ | ✓ | |
GPT-5.4GPT-4.1o3o4-mini | Native | ✓ | ✓ | |
Gemini 3.1 ProGemini 3 FlashGemini 2.5 Pro | Native | ✓ | ✓ | |
| OpenAI-Compatible | DeepSeekQwenOllamaGroqTogether AIvLLMLM Studiollama.cpp | API | ✓ / fallback | ✓ |
Note: For models without native tool calling, Copair falls back to prompt-based tool extraction — it still works, just with slightly less reliability.
Get Started
Install in seconds, configure in minutes
npm (Recommended)
Available on npmjs.com/package/@dugleelabs/copair
Homebrew (macOS/Linux)
Run Without Installing
Quick Configuration
Create a config file at ~/.copair/config.yml
version: 1
default_model: claude-sonnet
providers:
anthropic:
api_key: ${ANTHROPIC_API_KEY}
models:
claude-sonnet:
id: claude-sonnet-4-6-20260320More Features
Slash Commands
Custom Workflows
Multi-step YAML workflows combining prompts, shell commands, conditions, and branching. Build complex automation sequences.
Knowledge Base
COPAIR_KNOWLEDGE.md — project-specific facts auto-maintained by the agent and committed to git.
Token Tracking
Per-response and per-session cost tracking with per-model breakdown on exit. Know exactly what you are spending.