Request draft
Staff trigger an AI action from a specific screen such as Smart Insights, Appointments, Patients, Consultations, Pharmacy, or Billing.
AI Copilot
Unified request, queue, review workspace, approval, discard, and regeneration flow for every Operix AI surface.
RedactionQueued jobsReview workspaceROI
Workflow
How This Flow Lands In The Product
Build minimum context
Operix extracts only the smallest structured context needed for the capability.
Redact before egress
PHI is replaced with deterministic placeholders before outbound provider calls.
Queue and persist
Provider calls run in the queue and persist into ai_runs with prompt version, model, usage, latency, and safety flags.
Review and decide
Staff can approve, discard, copy, or regenerate the draft; stale drafts are flagged when the source changed.
Details
Product Surface
AI is now a coherent product layer instead of scattered buttons.
- Dashboard summary cards and clinic scorecard backlog signals
- Smart Insights daily, weekly, and executive briefs
- AI Run Log Review console
- AI Settings control plane
- Module-level request actions with freshness, redaction, and review state badges
- Source summary versus draft preview, reviewer comments, feedback tags, reviewer disposition and rubric metadata, exportable drafts, and regenerate lineage in the run log
- Optional handoff into communications or clipboard workflows without auto-send or auto-write
Details
Capabilities In v1
The current release deliberately favors high-utility drafts over open-ended chat.
- Smart Insights: daily brief, weekly brief, executive brief
- Appointments: reminder, no-show, reschedule, call script, bulk follow-up summary, queue pressure
- Patients: timeline summary and handoff checklist
- Consultations: SOAP, after-visit, follow-up, patient-friendly summary, referral letter
- Pharmacy: stock note, expiry action note, reorder rationale
- Billing: collections draft, anomaly explanation, payment plan draft
Out of scope in v1
No patient chatbot, no document analysis, no voice, and no autonomous clinical decisioning are shipped in this version.
Details
Provider Guardrails
Repeated provider failures now open a short-lived circuit breaker so staff are not stuck hammering a broken provider.
- AI drafting pauses temporarily after repeated failed runs in a short window.
- A successful provider health test clears the pause state.
- The queue contract stays the same: no AI output writes into records without a human action.
Details
Quality, ROI, Variants, And Regeneration
Buyer-facing AI proof is stronger when every capability shows both operational safety and practical return.
- Each capability now carries UI grouping, supported channel variants, default output mode, and conservative estimated minutes saved.
- Outreach drafts support sms, whatsapp, email, or call-note variants where the clinic workflow already exists.
- Reviewer disposition and rubric fields now persist alongside comments so quality review is not collapsed into a free-text note.
- AI usage snapshots track requests, approvals, stale rate, failures, latency, tokens, and estimated minutes saved without storing raw PHI.
- Regenerations create new runs with lineage instead of mutating prior history, which keeps review and audit intact.
- Clinic scorecard and Plan & Limits surfaces reuse non-PHI AI backlog and ROI signals so buyer demos can talk value before opening an individual run.
Coverage appendix
This page is tied back to routes, views, requests, tests, and implementation files in the generated docs coverage appendix.