v2.5

February 12, 2026

Richer human-in-the-loop controls for multi-agent teams

The human-in-the-loop (HITL) system for teams has been significantly expanded. New run requirements support tool confirmation, user input collection, and external tool execution, giving you fine-grained control over when and how humans interact with running teams.

This means teams can pause mid-execution to confirm a tool call, collect additional input from a user, or wait for an external action to complete before continuing. The result is safer, more controllable multi-agent workflows that keep humans in the loop exactly where it matters.

Details

  • Tool confirmation pauses before executing a tool and waits for human approval
  • User input collection requests information from a user mid-run
  • External tool execution allows a human or external system to perform a step and return the result
  • All controls work within team orchestration, not just single agents

Who this is for: Teams deploying multi-agent workflows in customer-facing, regulated, or high-stakes environments where human oversight at specific decision points is required.

Learn more in the HITL docs