Documentation
Learn how to install, configure, and get the most out of Copair.
Getting Started
Install Copair, configure your AI provider, and run your first coding session in minutes. Supports Claude, GPT-4o, Gemini, Llama, and more.
Model Switching
Switch between Claude, GPT-4o, Gemini, Llama, and any OpenAI-compatible model mid-session using the /model command — no restart required.
Context Persistence
Project-scoped sessions with auto-derived identifiers, automatic summarization at context limits, and /session commands to list, resume, or fork.
Git Integration
Automatic branch creation, meaningful commit messages, diff review at the approval prompt, and configurable co-author identity for every commit.
Custom Tools
Copair's 8 built-in tools — Read, Write, Edit, Bash, Git, Grep, Glob, WebSearch — plus a plugin system for extending with your own custom tools.
Permissions & Approval
Human-in-the-loop approval gate, allow-list YAML, always-ask and auto-approve modes, and session-scoped approval keys for safe autonomous operation.
Slash Commands
All built-in /commands documented, plus a guide to authoring custom commands with prompt templates, variable interpolation, and argument handling.
Workflows
Automate repeatable dev processes with YAML workflows — sequential steps, branching logic, shell commands, agent prompts, and loop-and-retry patterns.
Headless Mode
Run copair non-interactively for automation and benchmarking — a prompt in, a structured JSON result out, no TTY. Flags, exit codes, the result and event-stream JSON Schema contracts, approval policy, and the config-driven ablation toggles.
MCP Servers
Connect Copair to Model Context Protocol (MCP) servers to extend it with external tools, APIs, and data sources from the growing MCP ecosystem.
Token Tracking & Cost
Live token counts per request and session, per-model cost breakdown, and context window usage — so you always know exactly what you are spending.
Configuration Reference
Complete config.yaml reference covering provider credentials, agent identity, permissions, context window settings, workflow paths, and UI options.
Model Capabilities
What copair knows about each model out of the box — context window, max output, preferred format — plus how to override anything via model_overrides config.
Custom & Local Models
Run copair against a model it doesn't recognize — a fine-tune, a local GGUF, a private endpoint. Declare its tier with model_overrides and you're set.
Knowledge Base
COPAIR_KNOWLEDGE.md gives Copair persistent context about your codebase — architecture, conventions, and team preferences — shared across every session.
Security
Approval gate, PathGuard file system boundaries, secrets redaction, audit logging, prompt injection hardening, and MCP server environment isolation.