Skip to main content
Post a comment to guide a task’s behavior. Steering replaces the old approval model — instead of approve/deny, you guide the agent with natural language feedback.

​
Endpoint

POST /tasks/:id/steer

​
Request body

{
  "comment": "Focus more on open-source models next time",
  "runNow": true
}
FieldTypeRequiredDescription
commentstringYesYour feedback or instruction for the next run
runNowbooleanNoIf true, triggers an immediate run with this comment as context

​
Response

{
  "ok": true,
  "taskId": "abc123",
  "comment": "Focus more on open-source models next time",
  "runTriggered": true
}

​
How it works

  1. Your comment is stored in the task’s steering history
  2. On the next run, all steering comments are injected into the agent’s context
  3. The agent reads your feedback and adjusts its approach
  4. If runNow: true, a run starts immediately — otherwise, it waits for the next scheduled execution

​
Example

# Steer a task and trigger a new run
curl -X POST http://localhost:3001/tasks/abc123/steer \
  -H "Content-Type: application/json" \
  -d '{"comment": "Include pricing comparisons", "runNow": true}'