User Manual
Running Agents
Your agents — the Workforce — are specialized units that each handle a defined slice of work. This page covers starting a run, watching it on the Runs page, reading a run trace and its confidence score, and knowing when a run is flagged for your review.
The workforce
Each agent handles one kind of work and only escalates to your Inbox when a decision exceeds its authority. The agent types include the Orchestrator, Purchaser, Finance, HR, Legal, Customer Success, Sales, Security, Audit, Operations, Researcher, Summariser, AP Specialist, Procurement, Contract Reviewer, Payroll, and HR Query.
Starting a run
The quickest way to run your first agent is the Get started flow, which walks you through grabbing an API key, connecting an integration, and triggering a run.
- •
Pick an agent type
Choose the specialist for the job — for example Procurement Specialist, AP Specialist, HR Query, Contract Reviewer, or Inventory Monitor. - •
Describe the task
Give the agent a plain-language task in the task field (for example “Get 3 vendor quotes for 500 laptops”). - •
Run it
Start the run and watch its live status update in place.
◳ Screenshot
The 'Run your first agent' step of Get Started — an agent-type selector, a task field, a Run button, and the live run status.
The Runs page
Runs lists every agent run newest-first, with three header stats and a filterable table. The header status line shows how many runs are live and the total count.
Header stats
| Total Runs | Every run dispatched in your workspace. |
| Currently Running | Runs in progress right now (pulses blue when active). |
| Awaiting Triage | Runs flagged for a human decision. |
The table
Each row shows the agent type, the task, the status, and the confidence score. You can filter the list by status (for example running or failed) and by agent. Runs that need a decision are flagged for triage and also appear in your Inbox.
◳ Screenshot
The Runs page — the three stat cards (Total Runs, Currently Running, Awaiting Triage) above a table whose rows show agent type, task, status, and confidence, with status/agent filters.
Reading a run trace
Open any run to see its full trace. The run detail page shows the agent, a status badge, the duration and completion time, the original Task, and — when present — the Confidence Score, the structured Result, the agent's Reasoning, its Proposed Approach, Validator Notes, any error message, and the tokens used.
Confidence score
The confidence score is shown as a percentage with a plain-language label:
| 85% and above | High confidence |
| 65–84% | Moderate confidence |
| Below 65% | Low confidence |
requires_triage
When a run needs a human decision it is marked requires_triage. The detail page then shows a “Sent to Human Review” banner with the triage reason and a Review in Inbox → link. Low-confidence and out-of-policy runs are the usual triggers.
◳ Screenshot
A run detail page — status badge, the Task, a Confidence Score ring, the Result table, the Agent Reasoning quote, and a 'Sent to Human Review' triage banner linking to the Inbox.
Worked example: a low-confidence contract review
Example · When a run routes itself to your Inbox
You run a Contract Reviewer on a new supplier agreement. The agent extracts the key terms but returns a 58% confidence score because an indemnity clause is ambiguous. Because confidence is below the threshold, the run is marked requires_triage.
- On Runs, the row shows status Triage Required and the 58% score.
- On the run's detail page, the Sent to Human Review banner explains why.
- In your Inbox, the item appears with the agent's reasoning so you can decide.
You read the clause, make the call, and approve or reject from the Inbox — the run completes accordingly.