Start a run
- From Research: ask, for example
Run the Workflow "Security hunt" on payments-api. Check the inputs and budget on the card, then click Approve. - From the workflow: open its menu and click Run now.
- Automatically: add a trigger.
- From your AI assistant: connect MCP and ask it to start the workflow. This works for workflows that take no inputs or only findings.
Follow a run
Open a workflow and click Runs. Click a run to watch its steps update live. Click a step to see its output. To stop a run, click Stop run on the run page, or ask Research and approve the card. Results it already saved stay.Message a running agent
While a run is working, you can message one of its agents to guide it, for example to point it at a file or narrow what it checks. Messaging needs the Owner or Admin role by default.- Open the agent. On the run page, click a step, then View activity. From Research, click View agents on the run’s card.
- Type in the Message box under the agent’s activity and press Enter.
Pause and resume
A run you start from Research, Run now, the API, or the CLI pauses instead of ending:- At its budget. The run pauses when it spends its budget. If Research started it, a notice appears in that conversation.
- When your organization runs out of credits. If the run can pause, it pauses when it needs its next AI model call. Add credits before you click Resume.
- When you pause it. Click Pause on the run page. The run pauses at its next agent step and keeps its budget.
Approve tool calls
A tool’s Approval setting decides whether its calls wait for a person. You choose it when you create the workflow. See Approval. Approving tool calls needs the Owner or Admin role by default. On the run page, the Approvals tab lists the calls that wait for a decision. Ask first: waits for OK: the agent holds the call and keeps working on the rest of its task. Click Approve to make the call, or Deny to refuse it. You can also answer from Research: click View agents on the run’s card or notice. Ask later: runs once OK’d: Enclave saves the call, and the agent carries on. The run can finish before anyone decides. The person who started the run gets a notification. Click Approve to run the saved call, or Reject to drop it. A call no one decides eventually expires.Results
Open a run and click Outputs to see what it produced. Outputs appear as the run saves them, so you can open them before it finishes.- Saved outputs: the reports and files the run’s agents saved. Click Download to get any output.
- Findings: the findings the run created. Findings in repositories you can’t see are counted but not listed.