Integration · Projects and docs

Chat with GitHub.

Search and read repositories, issues, pull requests, and code on GitHub. Ask in plain English, and get the answer with its source linked.

Create free account

Free: 30 answers a day, plus 300 trial credits for every other model. No card.

Read by default Changes on approval No index

It reads

Repositories
Issues
Pull requests
Code
CI runs
Security alerts

It can change, with approval

Commit chat files

It never

Builds an index
Sees what you can't
Trains a model
Changes anything unasked

Try asking

Questions people ask GitHub.

Bot recipes

Bots that work with GitHub.

Start one from a recipe written for any tool of its kind, pick GitHub, and it works as you, asking before it changes anything unless you let it.

Issue triage
Labels new issues in your issue tracker from your set, links duplicates, asks for what's missing, and otherwise stays quiet.

New issues Any issue tracker
Pull request reviewer
Checks each pull request against your contribution rules, asks Copilot to review the ones that deserve it, and closes what your scope rules out, saying why.

GitHub repository events
Issue fixer on a computer
Turns an issue a maintainer labels into a draft pull request, running the tests on a computer the team lends it.

GitHub repository events
Answers from the code
Answers questions in issues and discussions from the repository's own code and docs, Copilot style.

GitHub repository events
Small fix pull requests
Turns an issue a maintainer labels into a small draft pull request, written from a copy of the code in the chat.

GitHub repository events
Release notes drafter
Drafts release notes in the style of the last two releases when a tag or draft release appears.

GitHub repository events

All recipes

Read-only by default. Only your GitHub.

It sees what your GitHub account can see, and nothing more. Switch it to Read and change, and each change waits for your approval, shown exactly as it will run.

It reads through GitHub's own MCP server, only when you ask. None of it trains a model. More on the security page.

Disconnect any time.

In Chat with Work: Settings, Integrations, Disconnect. Its tokens and what it cached are deleted.
Or pause it: A paused integration stays connected, but the assistant can't use it until you resume it.
More than one account: Connect each one, and name it. The assistant knows which is which, and each account keeps its own access level.

Step by step in the help center: Connecting services, and Troubleshooting an integration if it stops working.

Works with the rest of your work