GitHub Copilot has introduced three notable updates: a generally available local sandbox for agents across the CLI, desktop app, and VS Code; the ability for the Copilot app to separate the GitHub account used for licensing from the account used to access repositories; and a new model‑discovery command in the CLI that surfaces both local Ollama models and cloud‑provided models. Together these changes tighten control over code‑generation agents, simplify model selection, and give operators tools to manage storage consumption.
Local Sandboxing for Agent Isolation
The sandbox limits an agent’s ability to read files, open network connections, and retrieve credentials. It is now the default in the Copilot CLI, the Copilot desktop application, and VS Code sessions that use the Agent Host. Because the feature is included at no extra cost, teams can enable it without budgeting for additional licenses. The practical implication is a reduced attack surface for any code‑generation process that runs on developer machines or CI runners, helping security engineers meet data‑handling policies.
Multi‑Account Licensing in the Copilot App
The Copilot app now supports distinct GitHub accounts for the Copilot subscription and for repository access. For example, an organization can provision a corporate Copilot license while developers continue to work from their personal or project‑specific accounts. This decoupling simplifies compliance with corporate licensing policies and avoids the need to grant repository‑level permissions to the account that holds the license.
CLI Model Discovery and Local Ollama Integration
Within the Copilot CLI, the /model command lists all available models, including those served by a local Ollama instance and the cloud models provided by GitHub Copilot. This eliminates the need to leave the terminal to check model availability, streamlining the workflow for AI engineers who experiment with on‑premise models alongside the hosted offering.
VS Code Session Management Improvements
The 1.141 release adds an Agents window that can display multiple agent sessions side‑by‑side in a grid, making it easier to compare outputs or run concurrent tasks. A new "Open worktree cleanup" option in chat reveals the disk usage of inactive session worktrees and lets users delete them selectively, directly addressing storage bloat on developer machines.
Related CloudNinjas coverage: AI engineering.
What This Means For Practitioners
Teams should enable local sandboxing wherever Copilot agents run to enforce isolation of file and network access. Review licensing configurations to ensure the appropriate account is used for billing versus repository access. Use the /model command to inventory local and cloud models, and benchmark any on‑premise model before adopting it in production pipelines. Finally, monitor VS Code worktree usage and schedule regular clean‑up to keep developer environments lean. By incorporating these steps, engineers can maintain tighter security controls, reduce operational friction, and better align Copilot usage with organizational policies.
