Trajectory Monitor Cloud
The trajectory monitor is a parallel reviewer that reads a session's trajectory — the sequence of tool calls, model turns and their outcomes — and judges whether the session is still doing what it was asked to do. It runs on the control plane, beside the session, never in front of it: the agent's request is answered whether or not the monitor has finished.
The three modes
| Mode | What happens on a bad verdict | Default |
|---|---|---|
PASSIVE | The verdict is recorded and an alert is raised for review. Nothing changes for the agent. | Yes — every workspace, unless changed |
ACTIVE | The verdict suggests a KILL for that session, which the loop governor then applies. This is one of the two opt-in paths by which a judged finding can escalate to enforcement — see Enforcement Actions. | No |
OFF | The monitor does not run for the workspace. | No |
The default is PASSIVE. An earlier design document (LLD 52) specified OFF; the platform has shipped with PASSIVE as the effective default since the monitor landed, and defaulting an unset value to OFF now would switch the monitor off platform-wide, so the code and this page say PASSIVE.
Changing the mode
Settings › Security › Trajectory Monitoring (the Monitor mode control), or the API:
curl -X PUT "$INTUTIC_CONTROL_PLANE_URL/api/v1/workspace/settings" \
-H "Authorization: Bearer $INTUTIC_API_KEY" \
-H "Content-Type: application/json" \
-d '{"trajectory_monitor_mode": "ACTIVE"}'ACTIVE is available on Enterprise plans. Turning it on does not change any verdict already recorded; it changes what the next bad verdict does.
What ACTIVE escalates, and what it does not
ACTIVE only ever turns a verdict the monitor has already reached into a KILL for that session. It does not add checks, lower thresholds, or touch any other session. A break-glass grant scoped to a rule or detector does not exempt a session from the monitor, because the monitor is not a rule the grant can name.