<10ms
Create and update latency
Open source · Apache-2.0 · Self-hosted
The Headless EHR, Built for AI Agents
Haste Health normalizes Epic, Cerner, HL7v2, and any FHIR-compatible system into one API, giving you a headless layer to power your apps and AI agents.
Haste Health
$ haste-health api search-type Patient "name=chen&_count=1"
{
"resourceType": "Bundle",
"type": "searchset",
"total": 1,
"entry": [{
"resource": {
"resourceType": "Patient",
"id": "xufn84vpa69b_998c1",
"name": [{ "family": "Chen", "given": ["Maya"] }]
}
}]
}Headless by design
Every Data Source In. Every Application Out.
Point EHRs, HL7v2 interfaces, and other FHIR servers at Haste Health. What comes out is one normalized FHIR R4 API that powers patient and provider apps, analytics, partner integrations, and AI agents alike.
Headless FHIR R4 APIFor builders
What You Don't Have to Build Yourself
The plumbing most healthcare startups end up writing from scratch, shipped as part of the platform.
Open Source, Self-Hosted
Apache-2.0 licensed. docker compose up runs the full stack locally, with pre-built binaries and container images for production. Own your infrastructure and your data from day one.
MCP-Native AI Access
14 tools generated live from your server's CapabilityStatement give agents structured search, read, write, and schema-discovery over FHIR data.
Ship Logic Without Forking the Server
Custom operations run as sandboxed TypeScript in an embedded Deno runtime, stored and versioned as FHIR OperationDefinition resources. Extend the platform, not a private fork of it.
Analytics Without a Data Warehouse
SQL-on-FHIR ViewDefinitions flatten resources into CSV, JSON, or NDJSON on demand, with a visual editor in the Admin App. Point a BI tool at clinical data without building an ETL pipeline.
Fine-Grained, Auditable Access
Attribute-based policies scope agents and users down to compartments and conditions. Every request can emit a real FHIR AuditEvent, and resource history is immutable at the storage layer.
Multi-Tenant From Day One
Tenant and Project scoping reach all the way down. Every query is filtered by tenant and project at the storage and search layers, so one deployment can serve every customer without re-architecting later.
See it
This Is What Headless Looks Like
An agent asks a plain-English question. Haste Health answers over MCP or the same auto-generated OpenAPI spec, with spec-conformant FHIR data and clear, actionable errors in the event of failure.
MCP · fhir_r4_search
Agent asks
"Find this patient's blood pressure readings from the last 30 days."
{
"tool": "fhir_r4_search",
"arguments": {
"resourceType": "Observation",
"code": "85354-9",
"date": "ge2026-07-19",
"patient": "xufn84vpa69b…998c1"
}
}Haste Health responds
{
"resourceType": "Bundle",
"type": "searchset",
"total": 1,
"entry": [{
"resource": {
"resourceType": "Observation",
"id": "bp-8f2a1c",
"meta": { "profile": [
"http://hl7.org/fhir/us/core/StructureDefinition/us-core-blood-pressure"
] },
"status": "final",
"category": [{ "coding": [
{ "system": ".../observation-category", "code": "vital-signs" }
] }],
"code": { "coding": [
{ "system": "http://loinc.org", "code": "85354-9" }
] },
"subject": { "reference": "Patient/xufn84vpa69b…998c1" },
"effectiveDateTime": "2026-08-12T09:14:00Z",
"component": [
{
"code": { "coding": [
{ "system": "http://loinc.org", "code": "8480-6" }
] },
"valueQuantity": {
"value": 128, "unit": "mmHg",
"system": "http://unitsofmeasure.org", "code": "mm[Hg]"
}
}
// + 1 more item: diastolic (LOINC 8462-4), same shape
]
}
}]
}Performance for Real Clinical Workloads
A Rust core built to keep latency low and throughput high without oversized infrastructure.
>25k/s
Writes per second, 10 threads
<50ms
Typical search response
<100MB
Memory footprint per instance
Benchmarked on a single machine with Postgres 18 and a Synthea-generated dataset, 10 threads.
Who it's for
Built for Teams Moving Fast on Healthcare Data
Digital Health Startups
Skip months of FHIR plumbing and compliance scaffolding. Self-host under Apache-2.0 and spend your runway on your product, not your data layer.
Provider & Care Platforms
Normalize Epic, Cerner, and HL7v2 feeds into one API for patient timelines, care coordination, and clinical workflows.
Payers & Risk Platforms
Support eligibility, prior authorization, and claims-adjacent workflows on FHIR-first APIs instead of brittle X12 glue code.
AI & Agent Builders
A headless data layer means agents get first-class, structured access — not a scraped UI or a bolted-on integration.
Secure by Default — Even for Autonomous Agents
TOTP-based MFA, argon2 password hashing, and CSRF-protected auth flows. Scoped OAuth clients for every agent. Attribute-based access policies. Immutable, versioned resource history. The controls a HIPAA review actually asks for, not a checkbox.