For your team / CMO and CNO

Protect clinical judgment for the moments that need it.

Your licensed staff spend a measurable share of the day on work that needs no license, and the work that does need one gets interrupted by it. The question is where the line sits and what happens the moment an agent hears something it should not handle.

Map a transition See safety and governance

Protect the lane that needs a license.

Routine work continues while clinical signals move immediately to a named licensed owner.

The boundary

What Rely does not do, written plainly

This is not a disclaimer at the bottom of a page. It is the operating rule that the agent is configured against and that navigators are trained on before they take live work.

No diagnosis, no treatment decisions, no medication changes.
No clinical interpretation of symptoms, results or history.
No independent safety decisions and no triage.
No clinical advice, in any channel, at any hour.
A novel case does not create a new rule. It enters review.

The safety path

What happens the moment the agent should stop

Configured stops

Clinical questions, safety-sensitive language, ambiguous intent, missing permissions, bad data, system failure and out-of-scope requests are stop conditions tested before launch, not behaviours we hope the model learns.

Owned transfer

A stop creates a named work state with context, a destination and an acceptance requirement. It does not create an untriaged transcript queue for your staff to sort through.

Fallback

If the intended path is unavailable, the configured fallback and the accountable owner stay visible on the case. Fallback activation is recorded as its own event so you can see how often it fires.

Bring the work pulling licensed staff away.

We will map where the line sits, what the agent stops on, and what we would have to measure for you to believe the result.

Map a transition See how we report results

The objection we hear most: will automation miss a safety signal or blur accountability?
It could, if the design allowed it, so the design does not depend on the model recognising danger. Stop conditions are enumerated, tested against routine and edge cases before launch, and re-tested when anything changes. When a stop fires the agent stops talking and a named owner receives the thread with context. Accountability stays with the licensed owner you designate, and the record shows who accepted it and when.
No configuration catches everything. That is why fallback, monitoring, sampling, trace review and an incident path exist, and why one novel case never silently retrains the agent. Bring your quality team to the test-case review and try to break it before launch. That is the review we want.
Clinical boundary
Rely does not supply clinical judgment. Navigator work is nonclinical and bound to an agreed scope, and diagnosis, treatment, medication decisions, clinical interpretation and urgent safety response remain with your designated licensed team. Coverage, permissions and service levels are defined for each program.