Escalation and safety-gate agent
Forces a handoff to a human when red flags appear.
Checked against clinicians
Recall on clinician-labelled emergencies; false-alarm cost.
Where a person decides
Customer clinicians own thresholds; counsel on device status.
Works on
Chat, voice, mental health.
On the roadmap. Tell us if you need it sooner. Reviewed .
The problem it solves
Some moments must never depend on a model's mood: chest pain, suicidal thoughts, a child with a fever. The system needs a hard stop that hands over to a person.
How it works
Detect red flags
Deterministic rules and tested detectors look for emergency and crisis content.
Force the handoff
When triggered, the reply is replaced by a safe message and a route to a human.
Log every trigger
Each trigger and its outcome is recorded for review.
An example, illustrative
'I can't stop thinking about ending it.' Bot continues the chat.
Gate triggers: crisis resources shown, human notified.
Your escalation policy and who receives handoffs.
No detector is perfect. We report recall and false alarms, and tune together with clinicians.
What it helps measure
Quick answers
What does the escalation gate do?
Deterministic detection of red-flag and crisis content and forced handoff.
How is the escalation gate checked against clinicians?
Recall on clinician-labelled emergencies; false-alarm cost.
Why not rely on the model?
A model can miss a cue on a bad run. A gate guarantees that known red flags always route to a person.
Does it block too much?
We measure false alarms and tune so clinicians aren't flooded.
What can the escalation gate not do?
Customer clinicians own thresholds; counsel on device status. No agent makes a clinical decision.