# GitHub Copilot CLI
GitHub Copilot CLI is [[GitHub]]'s terminal-native [[AI Agent Harness|agentic coding agent]], the CLI surface of [[GitHub Copilot]]. It went public preview in September 2025 and reached general availability in February 2026, positioning it as Microsoft/GitHub's direct answer to [[Claude Code]], [[Gemini CLI]], and [[Kimi CLI]].
It's a full agentic development environment in the terminal — plans, builds, reviews, and remembers across sessions without leaving the shell.
## Modes
- **Plan mode** (`Shift+Tab`) — analyzes the request, asks clarifying questions, produces a structured implementation plan before writing code
- **Autopilot mode** — fully autonomous execution; runs tools and commands without per-step approval
- **Interactive mode** — default conversational loop with approval prompts
## Sub-agent Delegation
Copilot CLI automatically delegates to specialized sub-agents and runs multiple in parallel:
- **Explore** — fast codebase analysis
- **Task** — running builds and tests
- **Code Review** — high-signal change review
- **Plan** — implementation planning
## Models
Multi-model routing across the latest from [[Anthropic]], [[OpenAI]], and Google: Claude Opus 4.6, Sonnet 4.6, Haiku 4.5, GPT-5.3-Codex, Gemini 3 Pro. Users can pick per task or let Copilot auto-route.
## Extensibility
- Built-in GitHub [[Model Context Protocol (MCP)|MCP]] server out of the box
- Custom MCP server support: `/mcp add|show|edit|disable|delete` (config in `~/.copilot/mcp-config.json`, overridable per project), plus experimental `/mcp search` which queries the GitHub MCP Registry, or the enterprise-configured private registry (see [[GitHub Copilot MCP Registry]])
- Plugins installable directly from GitHub repositories
- Shares the same agent runtime as [[GitHub Copilot SDK]] — same execution loop, exposed programmatically
## Sandboxing
Two isolation modes, both public preview (see [[GitHub Copilot Sandboxing]]): **local sandboxing** (`--experimental`) wraps agent commands in an OS-level, deny-by-default sandbox powered by [[Microsoft eXecution Container (MXC)]] (Seatbelt/Bubblewrap/ProcessContainer), inspectable via `/sandbox policy`; **cloud sandboxing** (`--cloud --experimental`) runs the whole session in an ephemeral GitHub-hosted Linux VM with resumable snapshots. Enterprises can enforce a minimum sandbox policy via [[GitHub Copilot Managed Settings]].
## Positioning vs Peers
Same agentic coding terminal category as [[Claude Code]], [[Gemini CLI]], [[Kimi CLI]], and [[OpenAI Codex]]. Differentiation:
- **Distribution** — bundled with existing Copilot subscriptions; no separate billing
- **GitHub integration** — first-class Issues, PRs, Actions, repo access
- **Multi-provider** — not locked to a single model vendor (unlike Claude Code or Gemini CLI)
## References
- Documentation: https://docs.github.com/en/copilot/concepts/agents/about-copilot-cli
- Source code: https://github.com/github/copilot-cli
- Features page: https://github.com/features/copilot/cli
- GA announcement: https://github.blog/changelog/2026-02-25-github-copilot-cli-is-now-generally-available/
- Agentic workflows blog: https://github.blog/ai-and-ml/github-copilot/power-agentic-workflows-in-your-terminal-with-github-copilot-cli/
- Beginners guide: https://github.blog/ai-and-ml/github-copilot/github-copilot-cli-for-beginners-getting-started-with-github-copilot-cli/
- Preview announcement: https://github.blog/changelog/2025-09-25-github-copilot-cli-is-now-in-public-preview
- Overview by [[Simon Willison]]: https://simonwillison.net/2025/Sep/25/github-copilot-cli
## Related
- [[GitHub Copilot]]
- [[GitHub Copilot App]]
- [[GitHub Copilot SDK]]
- [[GitHub]]
- [[Claude Code]]
- [[Gemini CLI]]
- [[Kimi CLI]]
- [[OpenAI Codex]]
- [[AI Agent Harness]]
- [[AI Agents]]
- [[Model Context Protocol (MCP)]]
- [[GitHub Copilot Sandboxing]]
- [[Microsoft eXecution Container (MXC)]]
- [[GitHub Copilot Managed Settings]]
- [[GitHub Copilot MCP Registry]]
- [[Awesome GitHub Copilot]]