GitHub to Migrate Web Copilot to Agent Sessions, Extending Conversation Retention from 28 Days to the Lifetime of the Account

GitHub has announced that, starting as early as September 28, 2026, it will merge Copilot Chat on github.com, GitHub Mobile Chat, and Copilot cloud agent under a single experience and policy framework. Technically, the web version will be fully migrated to the agent sessions originally used by cloud agent, while cloud agent will use GitHub Sandbox to execute tasks. This is more than an interface consolidation: chat data retention will change from 28 days to the entire lifetime of the account, and the unified experience will be enabled by default after launch. Organizations that opt out will lose access to Copilot on both the web and mobile, rather than merely disabling the cloud agent.

Another change that will directly affect CI costs is that Copilot code review’s `Default` effort will change from Lite to Balanced. GitHub documentation indicates that Balanced sends pull requests to models with stronger reasoning capabilities for longer analysis of complex logic, security-sensitive code, and cross-service changes. It will also consume more AI credits and may slightly increase GitHub Actions minutes. GitHub estimates that Lite consumes approximately $0.05 to $1 in AI credits per review, while Balanced consumes approximately $0.25 to $5. Actual costs will still vary depending on PR size, the model used, and custom instructions.

Code review’s agentic capabilities will use an Actions runner to collect full repository context and can hand recommendations to cloud agent to open a separate remediation PR. If GitHub-hosted runners are disabled by an organization, the system will fall back to a more limited review. Using larger runners will also incur additional Actions charges. Administrators can explicitly pin the effort to Lite at the organization or repository level; otherwise, existing `Default` settings will automatically switch to Balanced.

Teams should address two issues separately before the transition date: privacy and compliance personnel should assess whether long-term conversation retention could include source code, incident data, or credentials, while platform engineering teams should review the frequency of automated review triggers, AI credits budgets, and runner policies. GitHub has not yet published complete details about data export, deletion, or Sandbox isolation for the unified experience, so these areas will need to be verified again before the production rollout.