Documentation Index
Fetch the complete documentation index at: https://docs.medera.info/llms.txt
Use this file to discover all available pages before exploring further.
┌──────────────────────────────────────────────────────────────────┐
│ Canvas (Frontend) — AgentBuilderContext, AgentCanvas │
└──────────────────────────────────┬───────────────────────────────┘
│ REST + Socket.IO
▼
┌──────────────────────────────────────────────────────────────────┐
│ Backend /api/agent-workflows + /api/deployments + /api/care-* │
│ Auth · Tenant Context · PHI Audit · Subscription │
└─┬────────────────────────────────────────────────────────────────┘
│
├─ workflow-engine (Node.js)
│ ├─ WorkflowExecutor (471L) — _executeNode, _updateExecution, RLS
│ ├─ ExecutionContext (354L) — resolveVariable, executionTrace[]
│ └─ WorkflowQueue (285L) — Bull on Redis DB 2, 5 concurrent workers
│
└─ Care Orchestrator (AI Services, port 8000)
├─ /api/care-orchestrator/workflows · /agents · /tasks · /voice
├─ AgentType enum (INTAKE, CARE_COORDINATOR, REFERRAL_SPECIALIST, …)
├─ Workflow State Machine (30+ states across Referral / Rx / Lab / Care Plan)
└─ Event Bus (INTAKE_*, PA_*, REFERRAL_*, PRESCRIPTION_*, LAB_*)
Workflow execution
client → POST /api/agent-workflows/:id/execute
backend → insert into agent_workflow_executions (status=pending)
backend → WorkflowQueue.queueWorkflowExecution(executionId, ...)
│ (graceful sync fallback if Redis unavailable)
▼
worker → load execution + version snapshot
→ WorkflowExecutor.execute(executionId, organizationId)
├─ executeInOrgScope() — SET LOCAL app.current_organization_id
├─ find start node
└─ loop _executeNode(node, context)
├─ dispatch to NodeRunner.execute()
├─ on success → route to next nodes
├─ on { paused: true } → halt + persist state
└─ on error → route to error port (or re-throw)
worker → update agent_workflow_executions (status=completed|failed|paused)
worker → emit workflow:execution:* on Socket.IO
Real-time events (Socket.IO)
Emitted to the org room and per-execution room:
| Event | When |
|---|
workflow:execution:started | Execution claimed by a worker |
workflow:node:executing | _executeNode invoked |
workflow:node:completed | Node returned success() |
workflow:ehr-sync | ehr-sync node touched a connected EHR |
workflow:execution:paused | user-approval returned { paused: true } |
workflow:execution:completed | Reached end node |
workflow:execution:failed | Unhandled node error |
Tenant isolation
Background workers obtain a Postgres connection and immediately run SET LOCAL app.current_organization_id = '<org_id>' inside a transaction. RLS policies enforce that every read / write is filtered by organization_id.
Versioned execution
For published executions, the executor loads the immutable snapshot from agent_workflow_versions — draft edits made to the same workflow after the publish do not affect running executions.
