Copilot app adds multi-repo Projects and a computer use mode that clicks through desktop apps

GitHub, owned by Microsoft, has added Projects and a computer use mode to its Copilot desktop app. Both are in public preview, according to the company’s changelog post, and computer use can operate apps that have no API.

One home for work that spans repositories

GitHub describes a Project as a single place that holds the context, agents and results for work that does not fit in one repository or one session. A feature may touch several repositories, and a launch plan may need specs, designs and assets without any repository at all.

A user starts a Project by describing a goal and attaching material. GitHub lists repositories, design mocks, Slack threads and notes from a design meeting as examples. The user then picks the models and tools. Copilot runs as many agents as the job needs in the background and returns results, such as pull requests, for review. It checks in when it needs a decision.

Projects also support automations for repeatable tasks. GitHub’s own example is generating a new promo video each time a feature lands. For a narrow task in one repository, users can still open a session straight from that repository.

Material from linked services, Slack and Figma among them, can also feed the app when a user needs to investigate a request, write up a spec or iterate on a design while the code sits alongside. In chat it can generate and edit images from natural-language prompts.

Computer use reaches apps with no API

The more unusual feature is computer use, in public preview on macOS and Windows. GitHub says Copilot can read what an app shows, click controls, enter or edit text and move between apps, including those with no API, command-line interface or MCP integration. The post gives an expense-report workflow as an example.

Control is gated. Copilot asks for approval before it takes over an app, and organizations can switch the feature off through managed settings. Individual users turn it on under Settings > Computer Use with Enable Computer Use. On macOS, the Accessibility and Screen Recording permissions are also required.

The changelog says plainly that “Copilot Projects are now in public preview.” The app can be downloaded or updated, and GitHub points to its Community forum for discussion.

What the announcement leaves unsaid

The post does not say which Copilot plans include Projects or computer use, whether either carries extra cost, or how usage is metered. That matters because Copilot’s billing has been shifting toward metering. A same-day GitHub changelog entry says enterprises working with a GitHub or Microsoft account team can now get Copilot included with GitHub Enterprise, with usage metered against an enterprise budget from the first request. Chat, code review and Copilot in the cloud need available budget there, and a standalone seat for users without a GitHub Enterprise license costs $10 per user per month. Whether agent-heavy Projects draw on that budget is not stated, but a tool that runs many agents in parallel is the kind of workload that budget would govern.

Model cost is a second open question. In another same-day post, GitHub introduced Copilot Fusion, a model option that routes each task across several models. Users are billed for each model it uses at that model’s standard rate, and paid plans get a 30% discount for a limited time. GitHub reports that Fusion Max scored 60 on the Artificial Analysis Coding Agent Index, within two points of Claude Code with Opus 5.5 Max, at about half the average API cost per task. Those figures are GitHub’s own and were not independently checked here. The Projects post does not say whether Projects can use Fusion, though it does let users choose models.

Safety is the other gap. The post describes an approval prompt and an administrator switch for computer use, but gives no detail on how approval works over a long session, what is logged, or how Copilot handles sensitive data visible on screen. An agent that reads and clicks across apps without an API sees whatever the user sees, and the changelog does not address that. Nor does it say how accurate the feature is on complicated interfaces, and it offers no benchmark or failure rate.

The same questions apply to Projects, which can combine Slack threads and design files with code. The post does not explain what access permissions the connected tools require, or how Copilot keeps a Project’s context separate from other work.

Who can try it today

Anyone can download or update the Copilot app and look for the features, since both are public previews. Computer use is limited to macOS and Windows, with no mention of Linux. Everything else, including pricing and plan eligibility, has to be confirmed in the projects documentation and the computer use documentation that GitHub links from the post.

For developers, the practical change is scope. Until now a Copilot session generally centered on one repository, and this update lets a goal span repositories, chat tools and design files. Whether that holds up on real multi-repository work is something the changelog cannot answer, and the preview label suggests GitHub expects the behavior to keep changing.