GitHub Copilot CLI Is Now Generally Available

From Terminal Assistant to Full Agentic Environment

GitHub Copilot CLI has graduated from public preview to general availability, marking a significant milestone in bringing agentic AI development directly into the terminal. What started as a command-line assistant for generating shell commands and explaining terminal output has evolved into a comprehensive agentic development environment that can plan complex tasks, execute multistep workflows, edit files, run tests, and iterate autonomously — all without leaving the terminal.

The GA release is available to all paid Copilot subscribers across Pro, Pro+, Business, and Enterprise tiers. The public preview began in September 2025 and incorporated extensive feedback from developers who used the tool across a wide variety of workflows, from simple file operations to complex multi-repository refactoring tasks.

Autonomy Controls

Copilot CLI introduces granular autonomy controls that let developers decide how much latitude the agent has during task execution. At the most restrictive level, the agent asks for confirmation before every action — each file edit, each terminal command, each test run. At the most permissive level, the agent plans and executes entire workflows end-to-end, only pausing when it encounters an error it cannot resolve or when it needs clarification on requirements.

These autonomy levels are configurable per session and can be adjusted mid-task, giving developers the flexibility to start with high oversight on unfamiliar codebases and relax controls as they gain confidence in the agent's behavior. Enterprise administrators can set maximum autonomy levels as a policy control, ensuring that agents operating in production-adjacent environments always require human approval for destructive operations.

Model Selection

Copilot CLI supports multiple underlying models, giving developers the ability to choose the model best suited for their task. The model selector is available at session start and can be changed between tasks within the same session. Supported models include GPT-5.3-Codex for complex agentic tasks requiring deep reasoning, Claude for nuanced code understanding and generation, and faster models for simple command generation and explanation tasks.

Model selection is surfaced as a first-class configuration option rather than being buried in settings, reflecting GitHub's recognition that different development tasks have meaningfully different model requirements.

Extensibility

The CLI agent supports MCP (Model Context Protocol) servers, enabling developers to connect external tools and data sources directly into their terminal workflows. MCP servers can provide the agent with access to documentation, issue trackers, deployment systems, monitoring dashboards, and other development infrastructure — extending the agent's capabilities beyond what is available in the local filesystem and terminal environment.

MCP server configuration is managed through a JSON configuration file that supports both local (stdio) and remote (HTTP/SSE) server connections. Enterprise administrators can manage an MCP registry at the organization level, pre-approving specific servers and blocking others to maintain security and compliance requirements.

Session Management

Copilot CLI introduces persistent session management, allowing developers to pause and resume agentic tasks across terminal sessions. Session state — including conversation history, file modifications, and task progress — is preserved locally and can be resumed with a single command. This addresses one of the most common pain points from the preview period, where closing a terminal window meant losing all context from an in-progress task.

Sessions can also be shared as read-only transcripts, enabling developers to review what the agent did during an autonomous workflow, share debugging sessions with teammates, or create reproducible examples for bug reports.

Availability

Copilot CLI GA is available immediately for all Copilot Pro, Pro+, Business, and Enterprise subscribers. The tool can be installed via npm, Homebrew, or as a standalone binary for macOS, Linux, and Windows. Existing preview users will be automatically upgraded to the GA version with no configuration changes required.