Stanford Med · Unit 5B Medicine·Day shift·Mon · 9:42 AM

Cockpit Copilot

LIVE · Cockpit chat · model claude-sonnet-4-6
← All agents
LIVEHITL: requiredCockpit chat · model claude-sonnet-4-6

Cockpit Copilot

Anthropic tool-use loop (up to 6 turns) bound to 5 typed tools. The Copilot can read patient state and propose clinician pages, Voalte messages, patient outreach, or prior-auth drafts — but every proposal lands in the Pending Actions queue requiring a licensed reviewer's Approve click before any integration fires.

Total calls
0
Approved
0
Pending review
0
Total spend
$0.0000
0ms avg
Inputs
  • · Patient
  • · Barriers[]
  • · CopilotMessage[]
  • · hospitalId
  • · maxTurns (default 6)
Outputs
  • · assistantText
  • · proposedDrafts[] (PendingActionDraft)
  • · toolCallsLog[]
  • · inferenceId
  • · cost_usd
  • · latency_ms
Output schema
CopilotResult
Governing rule

Hard rule: every tool call lands in the Pending Actions queue. No integration fires without reviewer Approve.

Source files
  • · src/lib/ai/agents/cockpit-copilot.ts
  • · src/app/actions/run-cockpit-copilot.ts
Tool-use functions (5)
  • query_patient_stateRead current patient + barrier + task state
  • propose_clinician_pageDraft a TigerConnect role page (STAT/URGENT/ROUTINE)
  • propose_voalte_messageDraft a Voalte secure message to a named role
  • propose_patient_outreachDraft an SMS or Epic MyChart push to the patient/family
  • propose_pa_draftHand off to prior_auth_drafter for a specific drug
Recent inferences
Open full audit →
Loading…
Demo data · no PHI · mocked Epic + payer endpoints