GitHub Opens Copilot Projects for AI Agents Working Across Repositories

Anime-inspired illustration of a software developer coordinating several AI agents working on different software projects at a desk

GitHub introduced Copilot Projects in public preview at its Universe conference in San Francisco on October 28, giving developers a way to coordinate AI agents across multiple repositories and tasks from one desktop workspace. The release moves Copilot further beyond code suggestions toward managing entire software-development assignments.

The Copilot app update lets a developer define a goal, supply repositories, design files or meeting notes, and choose appropriate models and tools. Agents can then work in parallel and return results such as proposed pull requests for human review. The project keeps that context, work history and output in one place.

GitHub Universe 2026 opening keynote. The presentation covers Copilot Projects and automations beginning around 14:55. Video: GitHub/Microsoft.

Coordinating agents across repositories

A single feature may require changes to a user interface, a backend service and shared libraries. Copilot Projects is designed to organize those related jobs without requiring the developer to start a disconnected agent conversation for every repository. Users can still run focused sessions directly from an individual repository.

GitHub Copilot desktop app showing Projects navigation and workspace
The new Copilot Projects interface in the desktop app. Image: GitHub.

GitHub also introduced project automations for repeatable work. Its wider Universe rollout includes agent-driven review and tools intended to carry work from an initial request through testing and proposed changes. These features make oversight important: suggested code and pull requests still need to be evaluated before they are merged or deployed.

Copilot expands beyond code

The app can draw on connected services including Slack and Figma, helping users work with specifications and designs alongside their code. GitHub also added conversational image generation and editing inside the app, extending the software beyond conventional programming tasks.

GitHub Copilot app demonstrating image generation from a natural-language request
Image generation in the GitHub Copilot app. Image: GitHub.

A separate computer-use preview launched October 1, allowing Copilot to navigate desktop applications on Windows and macOS, including programs without a suitable API. The tool requests approval before taking control; administrators can disable the feature, and macOS users must grant accessibility and screen-recording permissions.

The launch builds on GitHub’s efforts to scale its infrastructure for AI-generated activity and follows its expanding selection of Copilot models. BitcoinVersus.Tech has also covered Copilot runtime engineering and Microsoft’s safeguards for AI agents.

Copilot Projects is available now as a public preview, but GitHub has not presented it as a guarantee that autonomous changes will be correct. Its practical value will depend on how well agents work across repositories, how clearly their decisions can be reviewed, and whether teams can control spending and access.

Editor’s Note: We volunteer daily to ensure the credibility of the information on this platform is Verifiably True. If you would like to support our work and help further secure the integrity of our research initiatives, please scan the QR code below to donate 5,000 sats.

Original BitcoinVersus.Tech Bitcoin donation QR code
Scan the QR code with a compatible Bitcoin wallet to donate 5,000 sats.

BitcoinVersus.tech is not a financial advisor. This media platform reports on financial subjects purely for informational purposes.

8 responses to “GitHub Opens Copilot Projects for AI Agents Working Across Repositories”

  1. […] changes provide context for projects such as a Rust-based TypeScript compiler and GitHub’s cross-repository Copilot agents. They suggest developers increasingly combine typed languages, automated code changes and human […]

    Like

  2. […] and enterprise policies the new controls support. Other recent work on Windows Search and desktop-connected software agents illustrates why explicit app boundaries are drawing […]

    Like

  3. […] A second feature, Continuous Memory, indexes coding-agent sessions alongside a map of the repository. Entire reports that its codebase graph reduced token usage by as much as 71% in internal testing. The service also integrates with Buildkite and Depot for build-and-test pipelines, creating an alternative to some GitHub-based agent workflows. […]

    Like

  4. […] feature differs from Copilot Fusion model routing and Copilot Projects, which address different aspects of model selection and multi-repository coordination. It also […]

    Like

  5. […] GitHub Copilot Projects’ cross-repository workflow, this public prototype is explicitly scoped to one repository. Its goal is not to introduce a new […]

    Like

  6. […] can coordinate work through Agentic Workflows, Copilot Projects and Copilot Fusion. GitHub also reports more than 30 million merged Dependabot pull requests, […]

    Like

  7. […] conversation. Harness-specific permissions, models and history remain separate. Earlier coverage of Copilot Projects coordinating work across repositories provides broader context for the growing need to manage multiple AI coding […]

    Like

  8. […] Related BitcoinVersus.Tech reporting on Copilot Fusion covers the orchestration concept, while Copilot Projects coverage examines multi-repository […]

    Like

Leave a Reply