SCRIMED AgentOS v1
A governed multi-agent operating layer for healthcare intelligence workflows.
AgentOS coordinates planner, router, specialist, TrustQA, memory, RBAC, sandbox, MCP connector, audit, and human approval systems while live clinical execution remains gated.
Operating boundary
AgentOS v1 plans and routes governed work. It does not execute live care.
SCRIMED AgentOS v1 orchestrates synthetic pilot and enterprise assessment workflows only. It does not autonomously diagnose, treat, submit payer transactions, contact patients, file clinical documentation, or execute live care operations.
Multi-agent orchestration
Planner, Router, TrustQA, and Governance agents define the execution spine.
planner
Planner Agent
AgentOS control plane. Translate enterprise objectives into bounded workflow plans with memory scope, evidence needs, approvals, and measurable outcomes.
task plan, approval graph, evidence request, measurement plan
- plan created
- scope constrained
- approval graph attached
- denied capabilities recorded
router
Router Agent
AgentOS control plane. Route tasks to specialist agents by workflow type, risk level, evidence requirement, and connector boundary.
agent assignment, sandbox selection, handoff trace, blocked action list
- agent routed
- sandbox selected
- permission evaluated
- handoff trace recorded
trustqa
TrustQA Verification Agent
Watchtower trust layer. Verify outputs against boundary, evidence, confidence, safety, source attribution, and human-review policy before release.
verification state, trust card disposition, escalation reason, release recommendation
- TrustQA check run
- confidence evaluated
- source attribution checked
- release blocked or approved
governance
Governance Agent
Atlas Trust Systems Stack. Maintain AI asset inventory, policy trace, exception logs, approval records, and shadow AI detection signals.
governance finding, risk queue, exception record, asset disposition
- asset registered
- shadow AI signal detected
- exception requested
- governance review completed
Healthcare service registry
AgentOS can orchestrate Sanar AI, DocuTwin, CareExplain, Ambient Scribe, TrialCore, PayerIQ, and future SCRIMED services.
Sanar AI
Clinical context summarization and review prompts for synthetic or approved assessment workflows.
- care-gap-detection, patient-onboarding, chronic-care-monitoring
- No diagnosis, treatment order, or patient-facing guidance without licensed clinician review.
DocuTwin
Draft-only documentation review with source trace, missing context, and clinician review state.
- ambient-documentation, documentation-quality-review
- No final note, EHR filing, or clinical claim insertion without clinician approval.
CareExplain
Evidence-backed explanation, provenance, confidence, and source attribution for recommendations.
- patient-education-review, recommendation-explanation, trust-card-generation
- Explanations are review support, not clinical advice or patient instruction.
Ambient Scribe
Conversation-to-draft-note workflows with review queues and no final-signature capability.
- ambient-documentation, source-trace-review
- Draft-only; no medical-record update without authorized human review.
TrialCore
Eligibility queue support with criteria trace, missing evidence, and research review requirements.
- trial-matching, research-operations
- No enrollment recommendation, outreach, or treatment recommendation without research review.
PayerIQ
Policy-aware prior authorization and claims workqueue support with payer-facing actions blocked.
- prior-authorization, claims-review, rcm-denial-risk
- No payer submission, coverage guarantee, final coding, or billing action without qualified review.
Specialist agent registry
Every specialist agent is scoped by permissions, inputs, outputs, audit events, guardrails, and human-review requirements.
planned
Prior Authorization Agent
Revenue and access workflows
Prepare reviewable prior authorization packets from structured clinical, coverage, and policy context.planned
Revenue Cycle Agent
Administrative automation
Identify documentation, coding, claim, and denial-workflow gaps without making final billing determinations.planned
Scheduling Agent
Care navigation
Coordinate scheduling options across clinical urgency, patient preference, eligibility, and operational capacity.staged
Trial Matching Agent
TrialCore
Map synthetic patient signals to trial eligibility questions and produce reviewable matching rationale.staged
Documentation Agent
DocuTwin
Produce reviewable clinical documentation drafts from structured context while preserving clinician authorship.foundation
Compliance Agent
Trust infrastructure
Check workflows against privacy, security, auditability, and human-review requirements before expansion.foundation
Interoperability Agent
Integration fabric
Map workflow needs to FHIR, HL7, claims, pricing, device, and future healthcare connector contracts.planned
Clinical Intelligence Agent
Clinical Copilot
Summarize clinical context and surface reviewable decision-support signals without replacing clinical judgment.planned
Research Agent
Research operations
Synthesize research context, evidence gaps, and cohort questions for human research teams.foundation
Governance Agent
Watchtower
Track model, workflow, policy, risk, and quality-gate changes across SCRIMED operating surfaces.planned
Supply Chain Agent
Operational intelligence
Detect supply, inventory, vendor, and care-continuity risks before they affect patient operations.Platform controls
RBAC, MCP connectors, memory, TrustQA, sandboxing, and audit are first-class operating systems.
4 roles
Least-privilege enterprise permissions for admins, clinicians, RCM reviewers, and runtime services.
Open memory fabric4 connector groups
Connector planning across EHR, payer, knowledge, and CRM systems with no-PHI pilot boundaries.
Inspect AgentOS API4 templates
Task execution is currently a synthetic planning engine with production requests denied by default.
Inspect task engineInteractive workspace
Generate synthetic AgentOS task plans, Atlas Trust Cards, audit preview, and observability packets.
Open evaluation workspace