GitHub Copilot Agents Get Session Streaming in Public Preview

On July 2, 2026, GitHub launched a public preview of session streaming for Copilot agents — a feature that for the first time gives enterprise organizations a standardized channel for monitoring AI agent activity as it happens.

What is session streaming and why does it matter?

Until now, organizations using GitHub Copilot agents operated with limited visibility: agents would complete tasks, and the organization would receive results without the ability to review the workflow in granular detail. Session streaming changes that dynamic — organizations can now access data on the prompts, model responses, and tool calls that the agent executes, across all Copilot clients within the enterprise environment.

This is particularly relevant for enterprise environments where managing AI usage is both a regulatory and a security question. Visibility into what agents do — which tools they call, which prompts they receive, which responses are generated — is the foundation for auditing, compliance, and internal reviews. European regulation, including the AI Act, requires organizations to document AI systems operating in semi-autonomous processes such as coding agents.

Two ways to access the data

Session streaming comes in two forms. The streaming endpoint lets organizations automatically forward session data to event collectors, SIEM tools, or Microsoft Purview — which is itself in public preview. The data stream flows automatically, without manual intervention, as soon as an agent session is underway.

The REST API endpoint GET /enterprises/{enterprise}/copilot/usage-records offers a pull-model alternative — organizations can fetch on demand the data for the last 48 hours of all Copilot agent activity in the organization. This approach suits batched analytics, periodic reporting, or integration into in-house BI tools.

The feature covers all Copilot clients: cloud agents, CLI, Visual Studio Code, Visual Studio, and partner IDEs. Regardless of which tool from the Copilot ecosystem users within the organization are using, session streaming captures everything in a single unified point.

Activation and availability

Administrators must navigate to the Copilot subsection within AI Controls in the enterprise settings and enable two options: Copilot Usage Records Streaming and Copilot Usage Records API. For both options, “Enable everywhere” must be selected. Without this activation, no data flows — the feature is off by default.

Availability is limited to GitHub Enterprise Cloud users with enterprise managed users (EMU). Organizations on other GitHub plans do not currently have access to this feature.

Context: from black box to audit trail

The growth in AI agent usage in software development has raised new organizational questions: who is spending what, which agents are doing what, and is it aligned with internal policies? Session streaming is GitHub’s answer to those demands. Instead of organizations building their own oversight mechanisms, they receive a standardized channel through which they can push data into their own SIEM systems such as Splunk, Microsoft Sentinel, or similar tools.

For security teams, this also means a new kind of responsibility in designing AI workflows: every tool call and every prompt the agent receives becomes an audited record. Coding agents are no longer “black boxes” — they become audit-capable entities whose activity can be tracked, archived, and analyzed.

Compatibility with Microsoft Purview (also in public preview) is particularly interesting for organizations already using Microsoft’s information and compliance ecosystem, as they can integrate Copilot data into existing compliance workflows without adding new tools.

Public preview as an invitation to test

Public preview means the feature is available for broader testing but is not yet in general availability (GA). GitHub is collecting feedback on integration approaches, data formats, and the needs of different enterprise environments.

For organizations that already have SIEM infrastructure, this is an ideal time to test the integration — while the feature is still in its formative stage and GitHub is actively accepting suggestions for adaptation. Session streaming is technically a single endpoint, but strategically it marks a turning point: visibility and governance of AI agents are becoming just as important as the agents themselves.