Get Started View RefactorAgent on GitHub View FHIR Platform on GitHub
Sovereign, Compliant & Agentic Healthcare AI

The Operating System for Healthcare AI Agents

OpenHealth Agents is the open-source, HIPAA-compliant framework engineered for autonomous clinical workflows. Connect Claude, Gemini, OpenAI, or local models to Epic, Cerner, and FHIR R4 β€” guarded by an active Guardian Layer that eliminates hallucinations and scrubs PHI in real time.

$ pip install openhealth-agents
HL7 FHIR R4 Native
HIPAA Compliant
Zero Data Retention
LangGraph Powered
agents.provider.scribe_agent Provider
Capabilities

Enterprise Healthcare AI Made Safe & Sovereign

Generic agent frameworks are dangerous in healthcare. OpenHealth Agents bridges the clinical trust gap with deterministic guardrails, zero-PHI transmission, and native FHIR interoperability.

The Guardian Layer πŸ›‘οΈ

An active perimeter wrapping every LLM call. Automatically redacts 18 HIPAA Safe Harbor identifiers, verifies clinical guidelines (CPGs), and executes dual-model hallucination criticism.

Native HL7 FHIR R4 Interop

Agents think in standardized FHIR resources (Patient, Observation, Condition, Encounter). Pre-built SMART on FHIR connectors for Epic, Cerner, Medplum, and HAPI.

LangGraph State Machine

Deterministic clinical control flow. Manage multi-agent collaboration with stateful patient memory, cyclical reasoning loops, and automatic fallback procedures.

Human-in-the-Loop (HITL)

High-stakes medical actions (prescriptions, claims denials, therapy modifications) halt into a cryptographic "Pending Signature" state, requiring clinician approval to execute.

Tamper-Evident Audit Trails

Every token, prompt, guideline check, and EHR mutation is cryptographically hashed with SHA-256 into an immutable append-only ledger for HIPAA & SOC2 compliance audits.

100% Sovereign & Open

Apache 2.0 licensed. Run entirely inside your private cloud (AWS, Azure, GCP) or on-premise hardware via vLLM or Ollama. No patient data ever touches third-party commercial clouds.

Built For

Engineered for the Entire Healthcare Ecosystem

Whether you are building healthcare software, deploying clinical agents into hospital workflows, or automating payer authorizations.

Developers & AI Engineers

Health AI Engineers

Build clinical applications with instant FHIR CRUD tools, typed Python schemas, and pre-configured safety guardrails right out of the box.

  • Python SDK & CLI with FHIR R4 types
  • Plug-and-play LangGraph templates
  • Model Context Protocol (MCP) server ready
Health Systems & Hospitals

Clinicians & Providers

Deploy ambient scribe, clinical summary, and chart review agents without compromising patient privacy or clinical judgment.

  • SMART on FHIR OAuth 2.0 integration
  • Clinical Practice Guideline (CPG) critic
  • Human-in-the-Loop approval workflows
Payers & Care Networks

Payers & Administrators

Automate prior authorization extraction, claims adjudication checks, and coverage verification with tamper-evident audit logs.

  • Da Vinci FHIR Prior-Auth implementation
  • CPT & ICD-10 medical coding cross-check
  • Cryptographic audit chain for CMS review
Architecture

The Layered Safety Architecture

How OpenHealth Agents isolates the reasoning brain from patient PHI and EHR systems using deterministic guardrails and cryptographic audits.

01. Interface Layer (Inbound & Outbound)
API / WebSockets / MCP
FastAPI & MCP Gateway

Exposes Model Context Protocol endpoints for Claude Desktop, ChatGPT, Gemini, and web apps.

Audio Stream Ingestion

Low-latency WebSocket audio pipe for ambient consultation and telemedicine transcription.

Clinician Review Dashboard

Streamlit & React-compatible interface for HITL approval requests and audit inspection.

↓ Protected by Cryptographic Isolation Barrier ↓
02. The Guardian Layer (Safety Perimeter)
Deterministic Guardrails
PII / PHI Redactor

Scans transcripts for 18 HIPAA identifiers using NER + regex and injects synthetic tokens before model inference.

Clinical Practice Guideline Critic

Secondary validation agent cross-references proposed interventions against AHA, ADA, and NICE guidelines.

Deterministic Action Sandbox

Circuit breaker that instantly halts unauthorized write mutations or black-box contraindications.

↓ Sanitized State Stream ↓
03. The Agentic Core (Reasoning Engine)
LangGraph Brain
LangGraph State Orchestrator

Multi-step cyclical graph coordinating provider, payer, and patient subagents.

Patient Vector Memory

Episodic longitudinal record retrieval with HIPAA-compliant encrypted embeddings.

Model Agnostic Inference

Connects to Azure OpenAI HIPAA, AWS Bedrock, GCP Vertex AI, or local on-prem vLLM / Ollama.

↓ Standardized FHIR R4 Action Calls ↓
04. Interoperability Layer (The Hands)
EHR & Tool Connectors
HL7 FHIR R4 Adapter

Complete CRUD support for 60+ FHIR resource types with strict JSON schema validation.

EHR Connectors

SMART on FHIR OAuth 2.0 implementations for Epic, Oracle Health (Cerner), and Medplum.

Clinical Calculators & Rx

Deterministic eGFR, CHAβ‚‚DSβ‚‚-VASc, MELD-Na, and drug-drug CYP450 interaction tables.

OpenHealth FHIR & MCP Stack

FastAPI FHIR R4 server + Streamable-HTTP FastMCP gateway. Exposes typed AI tools automatically.

Core Ecosystem Monorepo

OpenHealth FHIR Platform & MCP Gateway

Looking for our production FHIR R4 server and Model Context Protocol gateway? Explore OpenHealthAgents/fhir β€” a high-performance monorepo pairing a FastAPI + PostgreSQL 15 + Redis 7 FHIR R4 REST API with a Streamable-HTTP FastMCP Gateway that auto-generates typed AI tools directly from OpenAPI specifications with JWKS-validated JWT auth forwarding.

fhir-server (:8000) fhir-mcp (:8002) FastMCP Streamable-HTTP 20+ FHIR Resources PostgreSQL 15 + asyncpg
Catalog

Pre-Built Specialized Clinical Agents

Ready-to-deploy agents for clinical documentation, patient navigation, payer authorization, and clinical guideline oversight.

Provider πŸŽ™οΈ

Ambient Clinical Scribe

Captures doctor-patient audio, sanitizes PHI, generates structured SOAP documentation, and drafts FHIR Encounter resources with ICD-10 & CPT billing suggestions.

Encounter Observation Condition
Patient 🧭

Patient Navigator Agent

Empathetic, clinically bounded patient assistant for triage, pre-visit intake, medication reminder compliance, and scheduling follow-up visits via FHIR Appointment.

Appointment QuestionnaireResponse
Payer πŸ“‹

Claims & Prior-Auth Agent

Evaluates clinical notes against payer coverage policies, extracts required lab indicators, and prepares Da Vinci CRD / PAS prior authorization requests automatically.

Claim Coverage ClaimResponse
Safety βš–οΈ

Clinical Guideline Critic

Dual-model verification agent. Inspects proposed drug orders, lab workups, and diagnoses against authoritative Clinical Practice Guidelines (CPGs) before execution.

GuidanceResponse PlanDefinition
Provider πŸ”„

Longitudinal Record Synchronizer

Aggregates fragmented patient data across disparate EHR systems (Epic, Cerner, external labs) into a reconciled, deduplicated FHIR R4 episodic timeline.

Bundle MedicationStatement
Safety πŸ”’

Cryptographic Audit Sentinel

Records each agent invocation into a verifiable hash tree ledger with actor attribution, token counts, model checkpoint ID, and deterministic execution logs.

AuditEvent Provenance
Quick Start

Up and Running in Under 5 Minutes

Three simple steps to deploy sovereign, FHIR-native healthcare AI agents into your development stack.

1

Install Package

Install OpenHealth Agents via pip or clone the repository directly from GitHub.

$ pip install openhealth-agents
2

Configure EHR & Models

Connect your FHIR R4 sandbox (Epic, Cerner, or local HAPI) and private LLM endpoint.

$ openhealth init --ehr epic --sandbox
3

Run Scribe Agent Demo

Run the clinical ambient scribe with active Guardian PHI de-identification.

$ python -m agents.provider.scribe_agent.agent

            
          
Comparison

Why Healthcare Demands OpenHealth Agents

Generic agentic frameworks like LangChain, CrewAI, and AutoGen lack clinical guardrails, HIPAA compliance primitives, and native FHIR interoperability.

Capability / Feature OpenHealth Agents πŸ₯ Generic Frameworks (LangChain, AutoGen) Raw Foundation APIs (OpenAI, Anthropic)
Native HL7 FHIR R4 Primitives Built-in schemas for 60+ clinical resource types βœ“ Native & Validated βœ— Requires manual parsing βœ— Unstructured JSON strings
Active Guardian PHI De-identification 18 HIPAA Safe Harbor identifiers masked before LLM βœ“ Built-in Interceptor βœ— No healthcare guardrails βœ— Transmits raw tokens
Clinical Practice Guideline (CPG) Critic Validates orders against AHA/ADA/NICE medical standards βœ“ Dual-Agent Critic βœ— No clinical checks βœ— High hallucination risk
Human-in-the-Loop (HITL) Digital Signature Cryptographic lock on high-stakes clinical actions βœ“ Cryptographic Sign-off βœ— Basic boolean prompts βœ— Not available
Immutable Cryptographic Audit Trails SHA-256 Merkle chain recording every thought & tool call βœ“ HIPAA / SOC2 Ready βœ— Ephemeral console logs βœ— Proprietary provider logs
SMART on FHIR OAuth 2.0 Client Direct token refresh for Epic & Cerner integrations βœ“ Pre-built Connectors βœ— Requires external auth code βœ— Incompatible
100% Sovereign On-Prem Deployment Runs inside hospital firewall with zero cloud leakage βœ“ Fully Sovereign βœ“ Self-hostable βœ— Cloud SaaS only
Deployment & Support

Open Source First, Enterprise Hardened

Start building for free with our open-source framework under Apache 2.0. Scale with private cloud VPC deployment and enterprise clinical support when ready.

Open Source
Free
Self-hosted community edition under Apache 2.0
  • Full access to all 6 specialized agents
  • Native HL7 FHIR R4 connectors
  • The Guardian Layer (PII + CPG critic)
  • LangGraph state machine orchestrator
  • Local cryptographic audit logging
  • Community Discord & GitHub discussions
Get Started Free
Dedicated Enterprise
Custom
Fully managed sovereign clusters for hospital networks
  • Everything in Sovereign BYOC, plus:
  • Dedicated isolated Kubernetes clusters
  • 99.99% clinical uptime SLA
  • HITRUST & SOC 2 Type II compliance reports
  • 24/7/365 Clinical Engineering on-call
  • Tailored fine-tuning on institutional clinical data
Talk with Solutions Architect
Frequently Asked Questions

Everything You Need to Know

Have questions about compliance, EHR integration, latency, or open-source licensing?

How does OpenHealth Agents ensure HIPAA compliance?

OpenHealth Agents enforces HIPAA compliance at the architectural level. Before any prompt reaches an LLM, the Guardian Layer scans and redacts all 18 HIPAA Safe Harbor identifiers (names, MRNs, dates, locations, contact info) and substitutes them with synthetic cryptographic handles. Furthermore, when deployed on-premise or in your private VPC, patient data never traverses third-party servers, and every transaction is recorded in a tamper-evident audit ledger.

Can I connect OpenHealth Agents to Epic and Cerner EHRs?

Yes. OpenHealth Agents ships with pre-built SMART on FHIR OAuth 2.0 connectors specifically calibrated for Epic on FHIR and Oracle Health (Cerner) Millennium FHIR endpoints. You can run against the public Epic App Orchard sandbox for rapid prototyping, or inject your production institutional client certificates for real-world EHR deployment.

What is the latency overhead of the Guardian Layer?

The Guardian Layer is highly optimized and written in asynchronous Python with vectorized regex matching and compiled ONNX NER models. The complete PII redaction pipeline executes in under 20–35ms. Guideline verification runs concurrently during generation streaming, ensuring zero perceived lag for clinicians during ambient transcription.

Can we load our institution's custom Clinical Practice Guidelines (CPGs)?

Yes. OpenHealth Agents supports loading custom CPG rule sets formatted as FHIR PlanDefinitions, Clinical Quality Language (CQL), or markdown guideline documents. The Clinical Critic agent indexes these guidelines into local vector memory and strictly validates that agent plans adhere to your hospital’s formulary and care pathways.

How does OpenHealth Agents compare to LangCare?

While LangCare provides a great Model Context Protocol (MCP) server for FHIR data access, OpenHealth Agents is a complete sovereign operating system for healthcare AI. OpenHealth Agents provides the active Guardian Layer (PII redactor, hallucination critic, CPG validation), LangGraph multi-agent orchestration, Human-in-the-Loop cryptographic gates, and pre-built domain agents (Ambient Scribe, Patient Navigator, Claims & Prior-Auth). Both tools share the vision of open healthcare interoperability and can be used synergistically.

Build Sovereign Healthcare AI
Without Compromising Safety

Join leading hospitals, healthcare AI engineers, and digital health startups building HIPAA-compliant, FHIR-native autonomous agents.