Skip to Content
How-to guidesStart a run in the web

Start a run in the web and follow it

Launch a mission and watch your agent work in real time. Use a run for unattended work you check later: triage last night’s failing tests, review a pull request, or update dependencies across a repo. Where a space is for working side by side, a run works on its own and calls you only when it needs you. You start a run from a deputy, with a saved mission or a one-off prompt.

Before you start

  • Role: you can use the deputy.
  • A deputy with a manifest bound to it. See Create a deputy and a mission.
  • Available when your administrator enables runs and deputies.

Steps

Launch a mission

  1. In the left rail, open Deputies, then click the deputy’s name.
  2. In the Missions card, find the mission. It must be active. A draft mission can only do a Dry run, which renders the prompt without starting anything.
  3. Click Launch.
  4. Fill in the values the prompt asks for. Optional: change Time budget for this run, or click Add note to add Extra instructions for this run. The note applies to this run only.
  5. Click Launch.
  6. In the toast “Mission launched”, click Open run.

To launch a one-off prompt instead, click Launch ad-hoc mission on the deputy’s page. Optional: enter a Name. Enter the Prompt, which is sent as written, with no template. Click Launch.

If the mission is at its concurrency limit, the toast says “Queued”. The launch waits in line and starts when another run finishes. It shows under Waiting to start on the deputy page, with the title its caller gave it when there is one.

Follow and steer the run

  1. You are on the run page. If you left, open Runs in the left rail and click the run. The list shows the last 24 hours by default.
  2. Read the Activity panel. It shows the agent’s messages and tool calls as they happen. The stream state next to the title says live, reconnecting..., or finished.
  3. Watch the status badge in the header. A live run shows running. A run that waits for you shows awaiting human. A finished run shows succeeded, failed, timed out, or canceled. See Runs for every status.
  4. To steer it, type in the reply box under Activity to send a correction or new information. To drive it yourself, click Take over in the header. See Steer a live run.
  5. Watch the timeout pill (for example “12m left”). Click it to add time or to pause the timeout.
  6. Use the right-hand tabs for more: Details, Changes, Ports, Blocked, Resources, Costs, Audit.
  7. To stop the run, click Cancel run. It takes effect at once.

If the agent asks a question, a Needs you card is pinned at the top. See Answer a question from a run.

What you should see

  • A run page with the title, the status badge, and lines appearing in Activity. At the start you may see “The run’s space is being created…” and then “Waiting for the first line…”.
  • When the run ends, the badge changes to succeeded or another final status. An Output panel appears if the run produced output, and the Continue… button appears in the header.
  • The run in the Runs list with its status, trigger, and duration.

Next

Run a mission on a schedule or event: let the deputy start this work on its own, every night, or when another system sends a webhook.

Create a deputy and a missionStart a run from the CLI