Azure DevOps teams can connect existing repositories to the GitHub Copilot app without moving them to GitHub. The integration, announced as a public preview on November 17, 2026, supports local coding sessions and pull-request work; cloud sessions are not available for Azure repositories.
Check your accounts and model access
Install or update the app, install Git, then sign in with a GitHub account. You need either a Copilot plan or a configured model provider to make AI requests. The Microsoft account added for Azure DevOps does not replace this initial GitHub sign-in. GitHub’s quickstart explains the onboarding sequence.
For an employer-provided Copilot Business or Enterprise licence, check that the administrator permits the Copilot app. Its policy is separate from the CLI policy. Your Microsoft account must already have access to the Azure DevOps organization, project and repository; connecting it grants no additional Azure permissions.
Using your own model provider does not require a Copilot plan, but still requires GitHub sign-in. The provider setup is itself in public preview: open Settings → Model providers → Add provider and follow the configuration prompts. For company code, use the provider configuration your organization permits.
If some repositories are hosted on self-managed GitHub instead, follow the separate guide to connecting GitHub Enterprise Server to the Copilot app. That is a different account and repository connection.
Connect Azure DevOps and start a session
The Azure integration is off by default. Follow the documented connection sequence:
Open Settings, select Experimental, and enable Azure DevOps integration.
Select Accounts → Add account. Sign in with the Microsoft account that has access to the intended Azure DevOps repository.
In the sidebar, click the plus control beside Sessions, then Add Azure DevOps repository. Enter the organization name, choose the repository and follow the cloning prompts.
Click the Sessions plus control again and select the connected repository under Start session in. For the environment, use This computer.
Choose New worktree for a separate working copy, or Current checkout for the existing one. Select the base branch, session mode and model, then describe a small, reviewable change.
The session guide treats collaboration mode and tool permissions as separate choices. Interactive or Plan does not by itself mean every tool action will ask first. Select Ask every time if you want manual tool approvals; Assisted delegates some approval decisions to a model, while Approve all removes those prompts.
Inspect the changes and run the project’s normal checks before relying on the result. The Azure-specific documentation does not establish a minimum app version or explicitly confirm support for on-premises Azure DevOps Server or Team Foundation Version Control (TFVC) repositories; do not assume those environments are covered by this preview.
Choose who completes the pull request
A pull request is a proposed code change awaiting review and merge. In the app, use View PR to inspect the changes and checks. Fix asks the agent to address a review comment; review its follow-up edits too. The manual completion route uses Approve, then Complete and a merge confirmation, subject to your Azure permissions and repository policies.
Agent Merge is a separate, per-pull-request opt-in. Once enabled, it can address feedback, repair build or test failures, resolve conflicts and complete that pull request when Azure checks and policies pass—without another confirmation. Connecting the account does not automatically grant this authority.
If your process requires a person to make the final merge decision, leave Agent Merge off and complete the request manually. Also check session tool permissions: creating a pull request or replying to and resolving review threads can happen without another prompt under some settings. A final merge checkpoint and approval of earlier actions are different controls.
Check model costs before rolling out
Repository access does not pay for model usage. GitHub’s plan overview includes the app across Copilot plans, but Free has limited access and automatic model selection. Business is listed at US$19 per seat per month and Enterprise at US$39. Their monthly allowances contribute 1,900 and 3,900 AI credits per user respectively to a shared billing-entity pool—not separate individual balances.
Under Copilot’s token-based pricing, the model and amount of input, output and cached context determine usage; one AI credit equals US$0.01. A credit allowance is not a guaranteed number of sessions or completed tasks. An external provider has its own usage terms and costs.
For a concrete example, ten full-month Business seats cost US$190 and contribute 19,000 credits. If that pool has no other contributors and the team consumes a hypothetical 24,000 credits, permitted overage adds US$50, making the Copilot total US$240. This illustrative calculation assumes no budget blocks the usage and excludes taxes, negotiated rates, other services and separately billed providers.
Business and Enterprise allow additional paid usage by default. To stop spending beyond the included pool, an administrator can disable AI credits paid usage. A US$0 user-level budget is different: it blocks credit-consuming features immediately, even while pooled credits remain. GitHub’s organization billing guide explains both controls.
Separate connection failures from model problems
If a repository is missing or an Azure action is denied, check the signed-in Microsoft identity and its repository permissions. If the repository opens but AI work is unavailable, check the GitHub account’s app policy, model access and available credits, or the configured provider. Reconnecting Azure will not replenish a model budget.
A saved model choice may also be obsolete. The guide to replacing retired Copilot models covers that separate troubleshooting task.
Understand the data and access limits
Local execution does not mean every part of the workflow stays on the computer. GitHub’s Azure integration documentation says returned private code, pull-request details, comments and build-log excerpts can enter the selected model’s context. Chat and tool results are retained in local session history. Evaluate the model provider’s data terms as well as where the coding session runs.
There is no integration-specific setting to restrict access to selected repositories, and no separate integration audit log in the app. Azure permissions remain the access boundary, and actions appear in Azure under the signed-in user’s name. Choosing one repository for a session is therefore not equivalent to restricting the account to that repository.
For offboarding, signing out of the Microsoft account removes local authentication tokens but does not revoke Microsoft-side access. Use Microsoft Entra account controls for revocation as well; local sign-out and server-side access removal are separate steps.
