Give your agents a GitHub identity. Once connected, they can read repositories, search code, browse branches and commits, and — with write tools enabled — open issues, leave PR reviews, and comment on behalf of you. Every action on GitHub happens under your account.
One OAuth flow, scopes you can read. Connecting asks for repo and read:user. That's broad — it covers private repos and write access to code, issues, and PRs on any repo your GitHub account can reach. There is no narrower OAuth variant; if you want to cap what an agent can actually do, cap it at the tool allow-list on each AI Chat.
The token is yours. Visibility is the lever. You control which agents can discover the connection by setting visibility to private, owned drives, or all drives when you connect. But tools always execute with your GitHub token and your GitHub permissions — if a teammate's agent uses your connection, it's still your account on the other end.
Agents get a set of GitHub tools, split into read and write. Read covers repositories, file trees and contents, branches, commits, code search, issues (get, list, comments), and pull requests (get, list, diffs, reviews, inline review comments). Write covers creating and updating issues, posting comments, creating a PR review, and leaving inline PR review comments. Write tools are rate-limited more tightly than read tools.
Nothing renders inside PageSpace. GitHub is agent-only — there's no inline PR or issue view on a page, no webhook that files a GitHub issue as a PageSpace task. If you want to see what happened, you look at GitHub.
repo scope that grants full read and write on repos you can see. To keep an agent read-only, strip the write tools from its allow-list — there is no OAuth-level read-only variant.Search docs, blog posts, and more.