Chief AI One intelligence layer · every Careship product Reference
The idea in one line
Chief AI is the living layer between the patient journey and the model. It works out what matters for the task at hand, gives Claude exactly that, checks the result, and learns from what the human does with it.

Chief is not the chat box. The chat box, the Scribe, the patient profile, the call card and the dashboard alerts all use the same Chief layer.

Where people meet Chief

Chief AI layer · shared by every surface

hands a context packet toModel gateway→Claude (today) · other models later

What comes back

OutputBrief, note, answer, draft
ActionTask, message, push to EHR
VerifyDid it actually happen?
FeedbackAccepted · edited · rejected

Everything here goes back into the patient's memory, so the next touchpoint already knows.

Patient journey→Events→Chief AI→Relevant context→Claude→Output→Human / AI action→Verification→Outcome→Feedback↺
Seven questions Chief answers before any AI call
Three modes · every touchpoint uses at least one
Explicit

The user asks

"Prepare me for Rosa." "Why is she high priority?" Shows up in the Chief dock and on the patient profile.

Ambient

The workflow asks

Scribe, call card, message drafting, assessment reading. The user just gets a better workflow and never "uses AI".

Proactive

Chief notices

Open loops, overdue follow-ups, rising scores, missing tasks. Surfaced as dashboard cards and alerts, with authority rules attached.

Memory

Use-case library

Every Chief use case across Frontly, Visitly, Commly, Programly, CoCarely and Memberly, written up in one format. This is the build backlog. Built = live in the app today behind the Chief AI switch. Engine = exists in CoCarely's chief-intelligence Worker. Planned = not built yet.

Need the honest status, the capability × product matrix or a Markdown export? Open Status & matrix →

Skills & authority

What Chief can do, and who has the final say. The authority level belongs to the skill itself and is enforced by the platform, not just written into a prompt. The existing outcome libraries (Ella, Fia, Claire, Sera, Rivi) become entries in this registry. Shown: the skills this product uses.

Level 1 · AI decides

Low-risk and reversible. Chief acts and the human is notified. Example: send an appointment reminder.

Level 2 · AI recommends, human approves

Has real operational or care consequences. Nothing happens until someone approves. Example: push a note to the EHR.

Level 3 · AI advises, human decides

Clinical judgment. Chief lays out the evidence but makes no recommendation of its own. Example: interpreting an abnormal PHQ-9.

Reference page · synthetic data, no model is called here. The live version runs inside Visitly and Programly behind the Chief AI switch in the top bar. · Status & matrix