AI relapse monitoring for opioid recovery
that catches the rising risk window.

Embertide is positioned as a routing aid that sits on top of an existing care network — it does not displace sponsor relationships, clinical judgement, or licensed supervision. When a rising risk routes, the human at every step of the cascade retains override authority and writes the clinical decision back to the chart. Liability for the member's care path remains with the named clinician, sponsor, or treatment program. Embertide contributes a tamper-evident event log of every score, every notification, and every accepted or declined handoff, so any post-hoc review is reconstructible from the same artefacts the clinical team saw in real time.

Customers run the platform as an adjunct, not as the system of record, so the documentation your compliance and legal teams already rely on for sponsor and clinician coverage continues unchanged. Pilots typically adopt a short addendum naming Embertide as the notification layer; that is the only paperwork shift.

A daily risk model that pages too often is worse than no model at all — members disengage, sponsors mute the alerts, and the signal-to-noise ratio collapses. The platform is tuned for high specificity in opioid populations, not for raw sensitivity. The escalation threshold is configured per cohort by your clinical lead, not by a stock platform constant; peer and counselor overrides feed back into the model and sharpen the next threshold set. A score escalates only when the trajectory crosses the threshold AND the time-to-reach exceeds the response window, so transient noise is filtered before it can reach the member's phone.

In the cohorts we have instrumented so far, the average member sees a verified escalation roughly once per quarter — not once per week. If your network sees a different profile, throttle and threshold settings make that tunable from day one, and the override log gives the clinical lead a feedback loop to recalibrate without redeploying.

Check-in responses, voice and SMS transcripts, and peer-chat signals are encrypted at rest, scoped per member, and never sold, pooled, or shared outside the member's consented care team. Access follows least-privilege throughout the cascade: a peer sponsor sees the slice of context that lets them reach out effectively; a telehealth counselor sees the clinical slice; an aggregate outcomes dashboard for a treatment program sees outcomes, not individuals. Audit logs are full-fidelity and exportable to the network's compliance team on request.

Members consent at every integration point at the point of contact, and consent is revocable without losing the underlying recovery support. Withdrawal narrows who can be notified — it does not suspend the watchfulness itself, so a member who opts out of sponsor contact is still monitored for rising risk that 988 should pick up.

Clinical oversight sits with your clinical lead, not with the model. The platform surfaces an explainable risk trajectory and a recommended cascade order; your named clinical lead sets the escalation threshold per cohort, reviews override patterns weekly, and signs off on quarterly threshold-set updates. Embertide ships the audit trail, the override logging, and the FDA-trackable Software-as-a-Medical-Device envelope — the clinical judgement stays with your team at every step of the cascade. Crucially, no escalation moves from sponsor group to telehealth counselor or to 988 without the human at each step retaining the right to decline, redirect, or escalate the other way.

Pilots ship with a designated clinical-oversight contact on your side and a corresponding technical contact on ours. Weekly override-pattern reviews are the standing meeting; they keep the threshold calibrated to your cohort's real risk profile rather than to the cohort the model was originally trained on.

The 988 handoff is the safety net, not the first responder, and the platform is engineered around making that handoff reliable when it has to fire. Because peer, sponsor group, and telehealth counselor precede 988 in the cascade, a 988 call typically arrives only when the critical window genuinely closes without a pickup — so 988 specialists are not flooded with cold transfers. When the handoff does fire, the specialist receives the risk trajectory, the last 72 hours of check-ins, and the consented peer-chat context envelope, so the call is informed rather than anonymous.

Every 988 handoff is logged with the time-to-handshake, the context envelope that was sent, and the resolution that came back, so post-hoc review of the routing path is reconstructible. If your network operates in a jurisdiction with a different crisis line, the routing target is configurable per deployment — Embertide does not hard-code 988 anywhere the platform cannot redirect.