AI assists. Clinicians decide.

AI is used as a draft assistant inside controlled clinical and operational workflows.

• Summarize intake responses into a structured snapshot for a clinician to review.
• Draft clinician-to-patient communication that the clinician edits and approves before sending.
• Draft routine documentation like aftercare instructions, which a clinician then reviews and signs.
• Help our support team triage non-clinical questions, with handoff to a human for anything clinical.

These boundaries are enforced in code through our model policy and tool-allowlist gates.

• AI never makes autonomous clinical decisions or sends clinical communication without clinician approval.
• AI does not write to your chart without an explicit clinician approval step that is logged.
• PHI is never sent to a model that lacks an executed Business Associate Agreement.
• AI is never used to determine eligibility, deny coverage, or recommend medication without a licensed clinician in the loop.

Every AI call is recorded with a structured trace so we can answer 'who, what, when, why' for any model interaction.

• Each AI request is logged in AiInteraction with the prompt, the model used, and the redaction applied.
• Token usage is recorded in AiUsageLedger for cost attribution and quota enforcement.
• Model selection is governed by AiModelPolicy: only allow-listed providers may receive PHI-adjacent prompts.
• Site administrators can review AI activity for their organization at any time.