GitHub Copilot CLI

By the end of this lesson, you’ll have the new standalone Copilot CLI installed and working — a full agentic coding tool comparable to Claude Code and Antigravity, not just a helper for looking up commands.

What is GitHub Copilot CLI?

GitHub Copilot CLI is GitHub’s standalone agentic AI tool for the terminal. It is a complete rewrite — the old gh copilot extension (with its suggest and explain commands) has been retired. The new Copilot CLI is a fully autonomous agent that reads your codebase, edits files, runs commands, and iterates on errors, just like Claude Code and Antigravity.

Key features:

• Full agentic coding — reads, writes, and edits files across your project
• Auto model selection (GA July 2026) routes each task to a current model — the picker includes Claude Fable 5.1 / Opus 5 / Sonnet 5, GPT-5.6, and Gemini 3.8 Flash, with GPT-6 Astra arriving as the rollout reaches your account (in the CLI from v1.0.84). Paid plans can pin one with /model; Free and Student plans are Auto-only. Fable 5.1 and Astra start at Pro+, and on Business and Enterprise an admin has to switch Fable 5.1 on — its policy ships off
• Plan mode — /plan (or copilot --mode plan) outlines the work before execution
• Tool approval controls — fine-grained control over what the agent can do
• MCP support — connects to external tools and services
• Context management — /compact to free up context, /context to see token usage
• Deep GitHub integration — works seamlessly with Issues, PRs, and Actions
• Free tier available — works with Copilot Free (no credit card required)

Prerequisites

All you need is a GitHub account. If you don’t have one, create one at github.com — it’s free.

Copilot plans at a glance:

As of June 1, 2026 GitHub replaced the old per-request counters with AI Credits — a token-based system that applies to chat and agentic tasks across all plans. One credit is $0.01, consumed at each model’s API rate; code completions and Next Edit suggestions remain unlimited and unmetered on all paid plans, and pay-as-you-go overage is available if you run out.

GitHub doesn’t publish exact completion or AI-Credit allowances for Free or Student. The Business and Enterprise allowances shrank on 2026-09-01 because a promotional period ended, not because prices changed — the seat prices are unchanged. From 2026-10-01, every assigned Business or Enterprise seat is charged upfront, and going past the included usage can require additional payment before users continue. Overage is billed by default (1 credit = $0.01) unless your org sets a spending limit.

Sign-ups are open again. GitHub paused new Pro/Pro+/Student enrollment in spring 2026 and reopened it on June 17, 2026; Free and Student plans use Auto model selection only (since June 24, 2026). Copilot Business and Enterprise sign-ups, which had been closed, began reopening on September 1, 2026 — gradually, over the following weeks — for customers paying by credit card or PayPal — from 2026-10-01 every assigned seat there is charged upfront, and existing Business/Enterprise seats on that payment method pick up the same upfront per-seat charge starting with billing cycles from October 1, 2026. Check github.com/features/copilot/plans for the latest.

Education access:

• Students: the Copilot Student plan is free while you’re verified — an unpublished monthly AI-Credit allowance, Auto model only — and comes with the Student Developer Pack. You can upgrade to Pro and keep the pack
• Educators: Verified teachers may qualify for free Copilot benefits — check education.github.com for current terms
• Institutions: Free GitHub Enterprise is available through the GitHub Campus Program, but Enterprise access alone does not include Copilot licenses — those must be assigned separately

Installation

The new Copilot CLI is a standalone tool — it does not require the GitHub CLI (gh). Choose whichever install method works for your system:

Option A: npm (any platform — recommended if you have Node.js 22+)

npm install -g @github/copilot

Option B: Homebrew (macOS / Linux)

brew install --cask copilot-cli

Option C: winget (Windows native)

winget install GitHub.Copilot

Option D: Quick install script (macOS / Linux / WSL)

curl -fsSL https://gh.io/copilot-install | bash

Verify the installation

copilot --version

You should see a version number (something like 1.0.81). If you get “command not found,” restart your terminal and try again — the installer may have updated your PATH.

Authentication

First launch

copilot

On first launch, Copilot CLI will prompt you to authenticate. Run /login if prompted, or it will open the GitHub device flow automatically — you’ll see a URL and a one-time code to enter in your browser. This works even from remote servers over SSH.

Once authenticated, you’ll see the interactive prompt:

> copilot
Welcome to GitHub Copilot CLI!
Type your request, or /help for commands.

You’re in.

How Copilot CLI works

Copilot CLI operates in two modes: interactive and programmatic. Both are fully agentic — the tool reads your project, plans changes, edits files, and runs commands.

Interactive mode (default)

copilot

This opens a persistent session where you type requests, review the agent’s plan, and approve actions. It’s the mode you’ll use most often — just like using Claude Code or Antigravity interactively.

Programmatic mode

copilot -p "add error handling to all API routes"

The -p flag runs a single prompt non-interactively and exits. Useful for scripting, CI pipelines, or quick one-off tasks.

Your first interaction

Navigate to a project directory (or create one) and launch Copilot:

mkdir copilot-test && cd copilot-test
copilot

Try a simple request:

Create an index.html file with a "Hello World" page that has a centered heading and a dark background

Copilot will plan the task, create the file, and show you the result. Review the output and approve any file writes.

Plan mode

Type /plan (or launch with copilot --mode plan) to enter plan mode. In plan mode, Copilot creates a structured plan for your request and waits for your approval before executing. This is especially useful for multi-step tasks where you want to review the strategy first.

Key commands

Newer builds add /autopilot (run without approval prompts), /delegate — or an & prefix — to hand a task to Copilot’s cloud agent, which opens a PR, /fleet (run subagents in parallel), /tasks, /worktree, /skills, and /plugin (the old /plugins dashboard command was retired in v1.0.81 and split into /plugin, /mcp, and /skills). Run /help for the current full set — Copilot CLI’s command surface is expanding quickly.

Tool approval

Copilot CLI asks for permission before running commands or writing files. You control this with flags:

# Allow specific tools
copilot --allow-tool='shell(git)'

# Allow all tools (no approval prompts — use with caution; `--yolo` also allows all paths and URLs)
copilot --allow-all-tools

# Block specific tools — deny rules win even when everything else is allowed
copilot --allow-all-tools --deny-tool='shell(rm)' --deny-tool='shell(git push)'

# Put a ceiling on spend for an autonomous run
copilot --mode autopilot --max-ai-credits 50 -p "fix the failing tests"

Custom instructions

Create a .github/copilot-instructions.md file in your project root to give Copilot persistent context about your project’s conventions, architecture, and preferences. This works similarly to Claude Code’s CLAUDE.md and Antigravity’s GEMINI.md / AGENTS.md.

Copilot CLI vs the other tools

Antigravity note: Google retired the consumer Gemini CLI on 2026-06-18; individuals now use Antigravity CLI (agy). Enterprise Gemini Code Assist licenses and API keys keep Gemini CLI — see the Antigravity lesson.

All four are full agentic coding tools. The biggest differentiator is ecosystem fit: Copilot CLI integrates deeply with GitHub Issues/PRs/Actions, Claude Code excels at codebase reasoning, Codex CLI plugs into the OpenAI/Codex Cloud workflow, and Antigravity gives you Google’s models with a real free tier. There’s no “free Claude Code” tier — Claude Code requires Pro or higher.

Copilot Chat in VS Code

While you’re here, it’s worth knowing that Copilot Chat in VS Code now offers four modes:

• Ask — ask questions about your codebase
• Edit — make targeted edits to specific files
• Agent — autonomous multi-file coding (similar to using Copilot CLI)
• Plan — create and review structured plans before executing

These are complementary to the CLI — use VS Code’s Copilot Chat when you want visual context, and the CLI when you want speed and automation.

Verify it works

Paste this exact command to confirm everything is set up correctly:

copilot --version

You should see a version number printed. If you get an error, make sure copilot is installed and in your PATH (see installation steps above).

copilot -p "list all files in the current directory and describe what each one does"

Key Takeaways

• Copilot CLI is now a full agentic coding tool — the old gh copilot suggest/explain extension has been retired, replaced by a standalone copilot command
• Free tier with zero friction — Copilot Free gives CLI access to anyone with a GitHub account, no credit card required
• Interactive and programmatic modes — copilot for persistent sessions, copilot -p "prompt" for one-off tasks
• Tool approval keeps you safe — fine-grained --allow-tool and --deny-tool flags control what the agent can do
• Deep GitHub integration — the standout advantage over other AI CLI tools, with native understanding of Issues, PRs, and Actions