Jobs

A job is one conversation that ran: the prompt, the chain of tool calls it triggered, the result, and the trace. A scene collects many jobs over time — every prompt, refresh, and scheduled run is one. The scene's cells show the latest state; the job log shows how it got there.

What a job records

  • The prompt that started it
  • Every tool call the agent made — name, inputs, outputs, duration, cost
  • Any writes that needed approval, and the decision
  • The final response or artifact
  • A permalink that anyone with access can open

A job can start without a person

A job starts with a prompt — typed, scheduled, or fired by an event: a webhook, a calendar trigger, another scene's output. The agent picks tools, calls them, and can ask the user follow-up questions along the way.

A write pauses the job for approval

When the job reaches a write — a database, an external service, a system of record — the agent stops and asks. The user sees the exact diff and approves, rejects, or modifies it before the write commits. Job state is pendingreadyrunningwaiting (for approval or input) → completed, failed, or cancelled, and transitions stream in real time.

The trace covers debugging, sharing, and audit

When a job goes wrong, the trace shows which tool call failed and with what input. When a colleague asks what happened, the permalink shows them the same trace instead of screenshots. And every write is recorded with who, when, and why — Audit covers what gets logged where.

Jobs can run without a scene

One-off queries and ad-hoc tasks run as jobs too — same trace, same permalink, no persistent surface.

What's next

Updated 2026-07-20