GitHub Copilot CLI v1.0.76 Changelog
What changed in Copilot CLI v1.0.76, released 2026-07-30 β explained in plain English.
- You can now manage multiple chat sessions at once and easily reorder your queued messages.
- The tool gives you more control over plugins, AI models, and safety sandboxes.
- A bunch of annoying bugs are gone, like text you typed accidentally following you into the wrong chat.
Want a heads-up when GitHub Copilot CLI updates?
We email you the moment a new GitHub Copilot CLI version drops β plus plain-English release notes like these. No spam.
This site is protected by reCAPTCHA and the Google Privacy Policy and Terms of Service apply.
GitHub Copilot CLI is a tool that helps you write code and run commands right from your terminal. This update is all about giving you better control over your workflowβlike managing multiple projects at once, tweaking how the AI behaves, and fixing several glitches that made the tool feel clunky. Itβs a big step forward for power users who rely on the terminal to get things done quickly.
What's new
- Multi-session sidebar: You can now turn on an experimental sidebar to manage multiple chats at the same time. You can see their status, switch between them, and start new ones without losing your place.
- Message queue manager: If the AI is busy, you can line up (queue) your next messages. A new manager lets you reorder, edit, delete, or instantly send those queued messages.
- More control over add-ons: You can now turn features on or off individually, including plugins, instructions, AI agents, and language servers (tools that check your code as you type).
- New AI model: The tool now supports the "grok-4.5" model.
- Voice mode improvements: Voice mode will now automatically pause any music or media playing on your Mac or Windows computer while you talk, and resume it when you're done.
- Inline images: If you use a terminal that supports Kitty graphics (like Rio), the tool can now display images right inside the text window.
- Credit limit predictor: A new
/limits predictcommand looks at your past chats to guess a safe AI-credit limit for your current session. - Scheduled prompt counter: The bottom of your screen now shows how many scheduled prompts you have running.
- Timed status updates: You can now set custom status-bar commands to refresh automatically on a timer.
Improvements
- Faster file comparisons: Viewing large code changes (
/diff) is now faster and scrolls more smoothly. - Smarter web fetching: When the tool grabs data from the internet (
web_fetch), it now automatically follows redirects instead of failing. It will ask for permission if the redirect sends it to a new website. - Better autopilot mode: When you let the AI work on its own (autopilot), it now stays in that mode after finishing a task instead of reverting to manual mode.
- Smoother settings changes: If you turn your mouse settings on or off mid-chat, it updates instantly instead of requiring a restart.
- Safer sandboxes: Sandboxes (security barriers that stop the AI from touching sensitive files) are better at blocking tricky shortcuts on Mac and Linux. Enterprise admins can also lock down these security rules so users can't accidentally lower them.
- Smarter startup tips: The tool only suggests setting up instructions if you haven't already done it in your current project.
- Better subagent teamwork: The AI is now better at handing small tasks to "subagents" (helper AIs) to do things in parallel.
- Mid-chat model switching: If you change the AI model while it's typing, the tool waits until it finishes to make the switch, preventing glitches.
- Faster tool loading: MCP tools (external tools the AI can use) load much faster now thanks to better caching.
- Clearer update alerts: When an update downloads in the background, the notification is less alarming and simply suggests you restart to apply it.
Fixes
- Unsent text bug: Fixed an issue where text you typed but hadn't sent yet would follow you into a different chat session when you switched. It now stays where you typed it.
- "Holder terminated" crash: Fixed a bug where a chat would fail every turn with a "Holder terminated during creation" error after a helper AI finished its job.
- Broken hooks fixed: Fixed multiple issues with "hooks" (custom scripts that run automatically). If a hook returns bad data, it no longer corrupts your chat or injects the word "null" into the AI. It also now stops huge scripts from eating up all your computer's memory.
- Worktree directory bug: Fixed an issue where your working folder would suddenly snap back to the original folder shortly after you switched to a new one.
- Queue display fixes: The queued messages list no longer shows blank rows or incorrect counts, and pressing Ctrl+C now properly removes your newest queued message.
- Network proxy rules: Fixed a bug where a network proxy could bypass your rules and let the AI reach the internet when it shouldn't.
- Sandbox search prompts: Fixed an issue where the tool would ask you twice to bypass a security sandbox during a search.
Breaking changes
No breaking changes in this release.
How to upgrade
Update to GitHub Copilot CLI v1.0.76 by running:
npm install -g @github/copilot
This is our plain-language summary. Read the complete, official notes on the GitHub Copilot CLI official changelog β.
