Jobs

A Job is the execution record of any autonomous work in River. Every Agent run, every Workflow run, every scheduled task, every background AI analysis creates a Job. Jobs are River's way of letting you see what was done, what it produced, and what (if anything) went wrong.

Job lifecycle#

Every Job moves through a small set of states:

queued → running → succeeded
                 ↘ failed
                 ↘ cancelled

• Queued — the work is scheduled to start
• Running — execution is in progress; logs stream in real time
• Succeeded — finished cleanly
• Failed — something went wrong; the error is recorded
• Cancelled — you (or the system) stopped it before it finished

What a Job records#

Each Job is a first-class entity with:

• The Agent / Workflow / Schedule that created it
• Start and end timestamps and duration
• All inputs the Job received
• Every tool the Agent called and what each returned
• Every node the Workflow ran and what each produced
• All outputs the Job wrote (drafts, docs, entities)
• A natural-language summary of what happened
• Any errors with stack traces

You can replay a Job, fork its inputs to debug, or open the entities it produced. Nothing is hidden.

Where you see Jobs#

Three places, depending on what you want:

• Agent / Workflow detail view — Job history for that one specialist
• AutomationsApp → Runs tab — every Job across your Dataspace, filterable
• Activity feed — Jobs interleaved with human edits, in chronological order

The Runs tab is the cleanest place to triage. Filter by status, time, and Agent, and you'll see all of today's autonomous work in one list.

Cancelling a Job#

A Job in queued or running can be cancelled at any time. Open it and click Cancel. The Agent stops mid-loop and writes a partial-result record so you can see what it had done before stopping.

Why Jobs matter#

The gap between an "AI assistant" and an "AI agent" is whether you can actually see what it did. Jobs are River's answer:

• Every autonomous action is auditable.
• Every failure has a record.
• Every output is traceable to the Agent or Workflow that produced it.
• You can review a week of autonomous work in one filtered view.

This is what makes it safe to give Agents real responsibility: not blind trust, but traceable trust.

Tips#

• Set up a weekly habit of reviewing the Runs tab — patterns of failed Jobs are how you find Agents that need tuning.
• A Job that consistently runs longer than you'd expect is a signal the Agent has too many tools or too vague a persona.
• The Job summary is AI-generated. If it's vague, the Agent's run was probably also vague — tighten the persona.