GitHub combines Copilot and notifications in its Teams app

One app, two migration paths

GitHub combined Copilot and repository notifications in its Microsoft Teams app on October 23, giving teams a single place to receive updates and ask an agent to investigate them. The change also starts a migration: the legacy GitHub app will retire on November 16, 2026.

The upgrade path depends on the app already installed. Users of GitHub notifications should update it; GitHub says existing notification capabilities continue working. Users of the separate legacy GitHub app must install the consolidated app to retain Copilot in Teams. That older app will be renamed GitHub (deprecated).

Source 1

What the consolidation changes

Copilot could already act on Teams conversations. GitHub's October 15 update described creating issues, making code changes and investigating bugs across several repositories within one conversation. Agent-driven coding in Teams therefore predates the October 23 consolidation, which brings that capability together with notifications.

The combined workflow starts with information a team already receives. GitHub's usage guide says issue and pull-request notifications form threads whose parent cards show status, assignees, reviewers and checks. Participants can then ask Copilot to investigate, plan or implement work, while colleagues add context and correct assumptions.

For example, a team discussing a reported defect could ask for an investigation before requesting a code change, then review the resulting pull request. This is an illustration of the documented workflow, not a test we conducted. Its practical value is keeping the discussion attached to the work while the agent runs asynchronously.

Work can continue on GitHub after the discussion. The cloud-agent guide documents requesting changes through a pull-request comment and checking the timeline for a started-work event. Review remains attached to the proposed code, with merging a separate step from the Teams conversation.

Source 1Source 2Source 5Source 10

Paid access also requires enabled cloud sandboxes

GitHub's integration documentation lists paid Copilot access, and the broader setup guide requires a Teams workspace where app installation is permitted. Both guides still label the integration a public preview.

Cloud sandboxes must also be enabled. For organization-provided access, GitHub says cloud sandboxing is disabled by default and an organization or enterprise owner must enable it. The task runs remotely in a temporary, isolated Linux environment hosted by GitHub.

These requirements make installation only one part of adoption. An employee with a paid seat and a visible Teams app may still need an organization or enterprise owner to enable cloud sandboxes before the workflow is usable.

Source 3Source 4Source 7

The thread becomes input, and identity changes the controls

GitHub says Copilot captures the entire thread and stores that context in generated artifacts. Starting a direct message can narrow what is supplied. That makes the conversation itself part of the coding input: unrelated or sensitive details in a shared thread matter beyond the immediate chat.

In direct messages, actions use the linked personal account's permissions. In a shared thread or channel, artifacts use the app's identity. Only repository writers can trigger changes; other participants can contribute input. Workspace guests and repository outside collaborators cannot start or steer sessions.

The usage guide adds a merging consequence: with repository rulesets, app-authored pull requests require one additional approval by default when the repository already requires at least one. Teams collaboration therefore does not erase the repository's review process.

The permissions reference also makes clear that this is an integration capable of changing repository content. Beyond reading code and repository information, Copilot requires content-write and workflow read/write permissions to open pull requests and initiate sessions.

Source 4Source 5Source 6

The subscription is only part of the cost

GitHub lists individual Copilot Pro at $10 a month, with agent interactions consuming AI credits. Cloud execution has separate meters: $0.000024 per compute second, $0.000003 per allocated GiB-second of memory and $0.005 per GiB-month of snapshot storage. A plan's included AI allowance is therefore not an all-in price for this workflow.

Compute and memory metering stop when the sandbox is stopped, but its saved snapshot continues to incur storage charges until deletion. The billing guide still labels sandboxes a public preview while dating its $10 monthly trial entitlement only through July 2026. That expired offer does not establish free October usage.

Administrators can set sandbox product or SKU budgets and enable blocking when a limit is reached. GitHub explicitly excludes sandboxes from its Bundled AI credits budget type. Teams deployment thus brings two distinct spending controls to configure: the agent's AI allowance and the environment in which it works.

Source 7Source 8Source 9