Skip to main content
You can run an agent whenever you need it, and every run is captured so you can open, review, and follow up on it later. Runs produced by an agent are kept together in the agent’s own history, separate from your everyday chats.

Ways to run

Interactively

Start a run from the dashboard. Choose streaming to follow live progress and receive final content, or a standard run to receive the final answer when it’s ready.

Through the API

Start runs from your own product. See the agent lifecycle for the API view.
Every run produces a chat you can open — the full back-and-forth, tools used, and citations, exactly as it happened.

Run status

The runs list shows where each run stands:
StatusWhat it means
runningThe agent is actively working. You can stop it in progress.
completedThe run finished with a final answer.
failedThe run ended without a final answer. Any partial output is kept.
cancelledThe run was stopped before it finished.
If a run ends without producing a final answer, it’s marked failed — but any partial output is preserved, so you can see how far the agent got and pick up from there.

Stopping and continuing

  • Stop a run in progress at any time. Work done so far is kept with the run.
  • Continue an existing agent chat to keep going in the same context — this reuses the same run rather than starting a new one.
Each run pins its core agent version and settings, so you can attribute it to the configuration it started with even after deploying a newer version. Model output can vary, while skills and delegates resolve at run time. See Deploying & versioning.

Where run history lives

Agent run chats are kept separate from your normal chat list and are reached from the agent’s runs. This keeps automated and agent-driven work organized on its own, so your everyday conversations stay uncluttered — and gives every scheduled, API, and interactive run a single, predictable home.
Scheduled runs land in the same history as interactive ones, so the runs list is the one place to check on everything an agent has done. See Scheduling.

Next steps

Automate runs

Put the agent on a schedule for background runs.

See the API lifecycle

Run agents and read their output from your own product.