PREVIEW 0.2.0 · LOCAL-FIRST · ZERO TELEMETRY
Terminal-native AI for clinical workflows that need control.
One inspectable agent across an operator runtime with protected hybrid medical services — plans you can read, execution you can audit.
prior authappealscodingclaims explanationcare coordinationrecord importsdocumentationFHIRliterature searchstructured artifacts
$openmed login— install link shared with approved evaluators
openmed — clinical · session 0142localtls
❯Review this denied prior auth for CPT 97110 and draft an appeal.
PLAN · 4 STEPS · MODE: CLINICAL
01parse referral packetdeid.scan
02match coverage criterialcd.criteria
03evidence gap analysispubmed.search
04draft appeal + citationsartifact
▸deid.scanreferral_packet.pdf✓ 0 PHI spans
▸icd10.lookupM54.50✓ low back pain
▸lcd.criteriaL34049✓ 4 of 6 met · 2 gaps
▸pubmed.search"lumbar stabilization"✓ 3 citations
✓appeal_draft.md written· 6 evidence links
●awaiting reviewer approval — token required to finalize
❯█
ZERO TELEMETRY
No analytics, tracking, or background usage reporting built into the product.
EXPLICIT REVIEW
Draft preview, approval tokens, provenance, and reviewer metadata before finalization.
OPT-IN AUDIT TRAIL
Private run traces to your Hugging Face dataset, bucket, or local fallback. Fingerprint-first.
01 — RUNTIME BOUNDARY
Protected clinical services, inspectable runtime, explicit review before action.
OpenMed keeps the operator review loop visible, separates medical-service endpoints from the model-provider path, and routes sensitive or high-volume clinical processing to dedicated services when those paths are invoked.
Inspectable runtime
The TUI, sessions, plans, and generated artifacts remain visible and locally controlled.
Protected clinical services
Extraction, de-identification, terminology, LOINC, RxNorm, MedlinePlus, and HCC run on dedicated medical-service endpoints.
Performance & cost efficiency
High-volume unstructured clinical processing stays on dedicated service planes instead of consuming frontier-model context for every page.
Preview deployment
During preview, the native medical-service tier is provisioned as OpenMed-hosted infrastructure.
02 — DEPLOYMENT MODEL
Hybrid medical services, not a black-box healthcare prompt.
OpenMed separates the operator runtime, the model-provider path, and the medical service tier — visible, configurable, and suited to sensitive or high-volume clinical processing.
PLANE 01 · LOCALyour machine
Operator runtime
Sessions, plans, tool traces, provenance, and workflow artifacts stay on the machine running OpenMed.
PLANE 02 · PROTECTEDdedicated endpoint
Clinical extraction & de-identification
NER, PII detection, and de-identification run through a dedicated native service endpoint.
PLANE 03 · PROTECTEDdedicated endpoint
Terminology, coding & HCC
Semantic search, validation, crosswalks, drug lookup & safety, risk adjustment, and comorbidity scoring across ICD-10-CM/PCS, CPT, SNOMED CT, LOINC, RxNorm, openFDA, HCC, CCSR, Orphanet, GEMs, MedlinePlus, and PubMed — on a separate protected service plane.
PLANE 03 · DATASET COVERAGE
74,719
ICD-10-CM
FY2026 diagnosis codes
79,115
ICD-10-PCS
FY2026 inpatient procedures
19,598
CPT / HCPCS
Procedure codes
359,930
SNOMED CT
Clinical concepts
101,701
LOINC
Lab & clinical observations
RxNorm
Drug terminology
NLM RxNav live API
HCC V28
Risk adjustment
9 model segments
RxHCC + ESRD
Risk models
Part D & dialysis
530+
CCSR categories
AHRQ clinical grouper
55
Comorbidities
Charlson + Elixhauser
11,645
Rare diseases
Orphanet ICD-10 crosswalk
86K+
GEMs
ICD-9 ↔ ICD-10 mappings
openFDADrug safetyLabels, adverse events, recalls
semantic search · validation · crosswalk · drug lookup & safety · patient education · risk adjustment · clinical classification · comorbidity scoring · rare-disease mapping
03 — FOR WHOM
Designed for the people shipping it and the teams using it.
OpenMed serves infrastructure-minded adopters and clinical operators in the same workflow surface.
FOR BUILDERS
Ship a terminal-native medical workflow with visible control surfaces.
An installable, configurable agent runtime instead of a black-box assistant — plans, traces, and artifacts you can read.
- Install with a provisioned shell command
- Configure models, profiles, native services, and optional MCP
- Inspect plans, tool traces, and workflow outputs
- Bring your own MCP endpoints when org tools are needed
FOR CLINICAL TEAMS
One interface for prior auth, coding, documentation, and care coordination.
Start from pre-built skills and built-in agent modes. Review workflow previews before finalizing.
- Start from pre-built clinical skills
- Choose the right built-in agent mode for the job
- Review workflow previews before finalizing
- Keep outputs structured, exportable, and evidence-linked
04 — CAPABILITIES
Capabilities grounded in the current product.
no vapor — if it's listed, it shipsCLINICAL WORKFLOWS
cap.01Prior auth + appeals
Review requests against CPT / ICD-10 criteria, LCD/NCD guidance, and denial evidence gaps.
cap.02Inbox + discharge
Reviewer-safe triage and handoffs, follow-up tasks, and patient-facing drafts.
cap.03Coding + claims
ICD-10, HCC V28, RxHCC / ESRD, RAF, code-audit, and claims-explainer workflows with terminology validation.
cap.04Notes + SOAP
Visit transcripts to structured notes — vitals parsing, med review flags, ICD-10, E/M hints.
RECORDS & EVIDENCE
cap.05Record imports + summary
Apple Health, Health Connect, C-CDA, FHIR export, or labs — one consumer-summary path.
cap.06Structured FHIR diff
Extract entities, build FHIR bundles, compare against a baseline with visible diffs.
cap.07PubMed + terminology
Built-in ICD-10-CM/PCS, CPT, SNOMED CT, LOINC, RxNorm, openFDA, MedlinePlus, CCSR, Orphanet, GEMs, and PubMed tooling — with crosswalks, validation, and comorbidity scoring.
cap.08Hybrid medical services
Hosted extraction, de-id, terminology, and HCC in preview — portable boundary for later self-managed deployment.
CONTROL & TRUST
cap.09Draft, finalize, diff
Draft preview, approval-token finalize, provenance, and case-run history — not one-shot chat.
cap.10Private trace storage
Opt-in traces to a private Hugging Face dataset, bucket, or local directory. Fingerprint-first.
cap.11Bring your own MCP
Attach remote JSON-RPC MCP servers when you need organization-specific tools.
cap.12Mode boundaries
clinical / consumer / coordination / plan lanes choose the right execution and review boundary.
05 — QUICKSTART
Install, authenticate, ask.
Three steps from a fresh machine to a live workflow in the terminal — across clinical, consumer, or care-coordination lanes.
01
Install
OpenMed is currently in preview. Installation is shared directly with approved evaluators.
$ install command · shared on approval
02
Authenticate
Use Codex OAuth or provide an API key, depending on the SDK path you want.
$
openmed login03
Ask
Describe the workflow in natural language and inspect the plan, tool traces, and outputs as they appear.
$
openmed06 — DOCUMENTATION
Docs that behave like product onboarding.
Start at the route you need.OVERVIEW
Getting Started →
Install, authenticate, choose a lane, and run the first workflow.
PRODUCT
CLI Reference →
Commands, agent modes, flags, and workflow-facing behavior.
PRODUCT
Agent TUI →
The terminal interface, plans, sessions, and keyboard flows.
PRODUCT
Skills →
The built-in skill library and how to add your own local skills.
PRODUCT
Profiles →
Built-in config presets and how they affect current settings.
INTEGRATIONS
MCP Connectors →
Add remote MCP servers and expose them as namespaced tools.
INTEGRATIONS
Native Medical Services →
How extraction, de-id, terminology, RxNorm, and HCC endpoints deploy.
TRUST
Privacy & Security →
Runtime boundaries, PHI mode meaning, token storage, network paths.
TRUST
Audit Trail →
Private run traces in Hugging Face datasets, buckets, or local storage.
START IN THE TERMINAL
Start with the CLI and choose the lane you need.
Local sessions and artifacts, four built-in agent modes, configurable model and medical-service endpoints, optional MCP, and workflow outputs you can inspect before you depend on them. Consumer-import and care-coordination lanes stay local — no outbound send, no external writeback.