Skip to main content
See what the agent is doing and inspect its work while a job runs:
  1. Stream agent logs. Follow agent actions, commands, and results in the Trajectory tab as new steps arrive.
  2. Watch the agent’s directory. Open the Stream tab to browse sandbox files and generated artifacts in real time. It connects automatically after sandbox is ready.
Add --stream to your run. Currently supports Claude Code and Codex on Daytona, Modal, local Docker:
In another terminal:
Run the browser and viewer on the same machine. For Daytona, make credentials available to both Harbor and the viewer. Add --no-delete to keep the sandbox available after the run. Don’t forget to delete later.