ClinicDx puts a specialist-grade evidence based clinical AI in every consultation room. Built open-source, runs offline, and grounded in WHO and MSF guidelines.
ClinicDx delivers four capabilities: Clinical Decision Support, Medical Scribe, Lab Digitization, and Imaging Analysis in one offline stack. OpenMRS-native and FHIR R4 ready.
ClinicDx is a clinical-grade AI system designed for real-world care delivery, built around frontline workflows, safety requirements, and evidence-based practice.
Built on FHIR R4 and REST API standards. ClinicDx connects to existing health information infrastructure without proprietary connectors, enabling integration with national EMRs and referral networks.
Differential rankings, drug guidance, and clinical protocols are configurable to local disease burden, formulary, and care level. ClinicDx does not impose a single global model on diverse clinical environments.
The complete codebase is publicly available, auditable, and free to deploy. Ministries of Health and implementing partners retain full ownership with no licensing fees and no vendor dependency.
ClinicDx-v1: an integrated vision encoder, a voice scribe model, and a curated guidelines knowledge base. Deployable offline on commodity hardware.
Clinicians face compounding challenges at every point of care delivery.
Manual EMR entry is slow, error-prone, and often happens hours after care, losing clinical nuance.
Clinical officers make life-or-death decisions alone, with no specialist and no protocol library.
Paper lab results don't get digitized, interpreted in context, or flagged for critical values.
DICOM files go unread. 3.6 radiologists per million means most images are never reviewed.
Each capability summarized below.
A second opinion at the point of care. Differentials, treatment plans, dosing, and drug safety — grounded in WHO and MSF guidelines, every time.
Speak the encounter. ClinicDx writes it. Structured FHIR observations captured in real time — no typing, no transcription lag, no lost detail.
On-device DICOM and film interpretation — no PACS, no radiologist, no internet. Structured findings and referral guidance at the point of care.
Point your camera at a printed lab slip. ClinicDx extracts every value, flags critical abnormals, and interprets results against the full patient record.
ClinicDx ships as a self-contained Docker Compose stack. Three services communicate on an internal bridge network. No external API calls are made at any point. OpenMRS 3.x is the first supported EMR integration.
Three services on an internal Docker bridge network. Only the middleware API is exposed to the host.
clinicdx-v1-q8.gguf served by llama-server. 3.9 GB GGUF, GPU or CPU mode. Handles CDS reasoning, imaging analysis, document OCR, and audio decoding in a single binary.
27,860-chunk WHO and MSF guideline index. BM25 + semantic RRF hybrid retrieval with a 4-slot clinical intent reranker. Queried live during CDS generation. Sub-25ms search latency.
FastAPI service orchestrating the multi-turn ReAct KB tool-use loop and Scribe pipeline. Streams CDS output via SSE. Exposes a REST API any EMR can call at :8321.
OpenMRS 3.x is the first supported integration. The ClinicDx ESM microfrontend (@clinicdx/esm-clinicdx-app) adds four workspaces to the patient chart automatically after installation.
Clone the repository, copy .env.example to .env, and run make up. All model and KB artifacts (~6.6 GB) auto-download from HuggingFace on first start.
Add @clinicdx/esm-clinicdx-app to your OpenMRS 3.x frontend configuration. The CDS, Scribe, OCR, and Imaging workspaces appear in the patient chart action menu automatically.
In OpenMRS Admin, set System Administration → Advanced Settings → @clinicdx/esm-clinicdx-app: middlewareUrl to the engine host. No further configuration needed.
ClinicDx exposes a REST API compatible with any FHIR R4 system. Use CDS, Scribe, OCR, and Imaging without OpenMRS.
The engine API runs on http://your-host:8321 independently of any EMR. Start with make up, then call the endpoints directly from your integration layer.
/cds/generate_stream
Streaming SSE clinical decision support with KB tool-use
/scribe/process
Transcription text to structured FHIR R4 observations
/scribe/process_audio
Raw audio to structured observations via AudioProjector
/api/health
Full stack health check: model, KB, and middleware status
Multilingual phrase-based scribe converts spoken input to structured SOAP notes in seconds.
Patient data never leaves the facility. No cloud, no external API, no privacy exposure.
Structured differentials and discriminating features guide every clinical assessment.
No cloud. No external API. No privacy exposure. ClinicDx operates entirely within the four walls of your facility, clinical intelligence without surveillance.
All AI computation happens locally on facility hardware. No data ever leaves the machine.
Patient data stays inside your own infrastructure, whether OpenMRS, a custom FHIR server, or any configured EMR. No third-party cloud access.
Structured outputs and audit trails support national digital health governance frameworks.
No. ClinicDx is designed from the ground up to operate fully offline. All AI inference runs on local hardware at the facility. There is no dependency on external servers, cloud APIs, or internet connectivity of any kind.
ClinicDx runs on commodity hardware: a standard workstation or rugged laptop with a modern CPU. No GPU is required for CDS and scribe functionality. Imaging analysis benefits from a mid-range discrete GPU but can operate in CPU-only mode.
ClinicDx integrates natively with OpenMRS 3.x as a microfrontend ESM module (@clinicdx/esm-clinicdx-app). It also exposes a FHIR R4 REST API, enabling integration with other EMR platforms or standalone deployment without OpenMRS.
The scribe feature supports Swahili, Amharic, French, Hausa, and English out of the box. CDS outputs are available in English and French, with localization roadmap prioritized by deployment region.
Completely. Patient data never leaves the facility. ClinicDx performs all AI computation locally, with no external API calls, no cloud uploads, and no telemetry. Outputs are structured and auditable, meeting WHO and national health data governance standards.
Yes. The CDS knowledge base is curated and aligned to WHO and MSF clinical protocols, including WHO Essential Medicines List, WHO Malaria Guidelines, and IMCI protocols. Every CDS output includes citations so clinicians can verify the evidence basis.
Pilot ClinicDx in your facilities to reduce diagnostic uncertainty, preserve data sovereignty, and return clinician time to patient care.