GitHub has launched a GitHub Copilot desktop software for agent-based software program growth. The app is in technical preview for present Copilot Professional, Professional+, Enterprise and Enterprise customers.
The desktop app is designed as a central workspace for builders managing a number of AI agent classes without delay. It brings energetic classes, points, pull requests and background automations right into a single “My Work” view throughout linked repositories.
Every session runs in its personal git worktree, giving builders an remoted copy of a department whereas an agent investigates a bug, works on a backlog merchandise or responds to evaluate suggestions. The app manages these worktrees robotically, eradicating the necessity for guide department setup and cleanup.
GitHub has additionally added Agent Merge, a software that follows a pull request by means of checks, evaluations and merging. Builders can select whether or not Copilot ought to handle failing steady integration checks, reply to evaluate feedback, or full the merge as soon as set circumstances are met.
Management centre
GitHub introduced the launch as a response to the rising use of AI brokers in software program engineering and the administration challenges that include them. Builders are more and more transferring between prompts, planning, coding, evaluate and merging by means of AI-assisted workflows, however present instruments usually depart work unfold throughout a number of home windows and interfaces.
The corporate additionally pointed to rising exercise on its platform. Commits on GitHub have almost doubled 12 months on 12 months to greater than 1.4 billion a month, whereas GitHub Actions is dealing with greater than 2 billion minutes per week.
That progress is placing extra strain on evaluate and validation. One purpose of the brand new desktop expertise is to make agent work simpler to examine, so builders can see what was tried, what was validated and the place human intervention continues to be wanted.
A brand new characteristic referred to as canvases sits alongside the app. GitHub described them as shared work surfaces that may show a plan, pull request, browser session, terminal, deployment, dashboard or workflow state, permitting each the developer and the agent to replace the identical view.
The purpose is to maneuver agent output past chat threads right into a format the place actions and selections are seen in a single place. Builders can edit, reorder, approve or redirect work straight throughout the canvas.
Sandbox controls
GitHub can also be increasing how Copilot brokers run code by providing each native and cloud sandboxes. The aim is to let brokers take a look at and iterate in contained environments with out affecting manufacturing techniques.
Within the native model, Copilot runs in an remoted setting on the consumer’s machine with restricted entry to recordsdata, networks and system capabilities. Native sandbox insurance policies may be centrally configured and enforced.
Within the cloud possibility, every sandbox runs in an remoted non permanent Linux setting hosted by GitHub. Organisations can set their very own insurance policies, and customers can resume classes remotely from completely different gadgets.
Assessment adjustments
GitHub has additionally up to date Copilot code evaluate instruments to deal with the bigger variety of pull requests AI brokers might generate. Customers can now lengthen code evaluate by means of customized agent expertise, MCP server connections and configurable GitHub Actions workflows.
A medium-tier evaluate possibility has been launched, sending pull requests to what GitHub referred to as a higher-reasoning mannequin. Repository directors can assign both low or medium steerage ranges relying on a repository’s danger or significance.
GitHub has additionally added a “/security-review” ability for security-focused checks. A “/rubberduck” ability is now typically obtainable and may use a number of mannequin households to critique an implementation and determine points.
Assist for Copilot code evaluate has additionally been prolonged to Azure DevOps, giving customers one-click evaluate, inline feedback and repair ideas in that setting.
Broader ecosystem
Outdoors the desktop app, GitHub mentioned its Copilot SDK is now typically obtainable in Node.js/TypeScript, Python, Go, .NET, Rust and Java. The software program growth package offers builders entry to the identical runtime utilized by the Copilot app to construct inside instruments and customized brokers.
Copilot CLI has additionally been up to date with a redesigned terminal interface, voice enter utilizing on-device speech-to-text, and scheduled duties. Cloud automations can now let brokers run on a schedule, react to GitHub occasions, open points and depart feedback, with permission requested earlier than every write motion by default.
GitHub can also be increasing integration with exterior software program distributors by means of partner-built agent apps embedded in GitHub. Named companions embrace LaunchDarkly, Vivid, Amplitude, Sonar, Endor Labs, Octopus Deploy, Packfiles, PagerDuty and Miro.
One buyer, Avanade, mentioned the product offers engineering groups a single place to direct a number of agent-led duties.
“The GitHub Copilot app is the most recent in a line of AI tooling from GitHub that’s reworking our enterprise. Shifting past AI help, the app has supplied a much-needed management middle for agentic growth. Our ahead deployed engineers can dispatch a cohort of brokers and handle a number of initiatives, all from one location. Easy accessibility to plans and autopilots with the flexibility to run interactive classes or step into code the place wanted,” mentioned David Jobling, Grasp Know-how Architect, Head of Know-how & Supply Futures, World Solutioning & Supply, Avanade.