Documentation

Open Medical Scribe

Open-source AI medical scribe — real-time transcription and structured note generation with full local privacy.

An open-source AI medical scribe that transcribes clinical encounters in real-time and generates structured medical notes. Runs fully local for maximum privacy — no patient data ever leaves your machine.

Perfect for: Clinicians, healthtech developers, medical NLP researchers.

Key Features

Real-time transcription

Whisper ONNX runs locally with Swedish-optimized KB Whisper model. Streaming support with live partial results.

Structured notes

Generates SOAP, H&P, Progress, DAP, Procedure, and Swedish Journalanteckning notes from transcripts.

Pluggable providers

Swap transcription and LLM backends with a single config change — local or cloud.

Desktop app

Electron app available in Light (~1 GB) and Full (~1.9 GB with bundled models) builds for macOS.

CLI tools

scribe-transcribe and scribe-note for scripting and batch workflows.

Web UI

Browser-based interface with settings page for all providers and configuration.

Privacy-first

Entire pipeline runs on your own hardware in local mode. No cloud required.

Swedish support

Optimized for Swedish clinical encounters with KB Whisper and Journalanteckning format.

Supported Providers

Transcription

Provider	Type	Notes
Whisper ONNX	Local	Default. Swedish-optimized KB Whisper model, runs in-process
faster-whisper	Local	Python-based, high performance
whisper.cpp	Local	C++ implementation
OpenAI Whisper	Cloud	OpenAI API
Deepgram	Cloud	Deepgram Nova
Google Cloud Speech	Cloud	Google Speech-to-Text
Berget AI	Cloud	EU sovereign cloud, Swedish Whisper

Note Generation

Provider	Type	Notes
Ollama	Local	Any local LLM (Llama 3.1, Mistral, etc.)
OpenAI	Cloud	GPT-4o and other models
Anthropic Claude	Cloud	Claude 3.5 Sonnet and other models
Google Gemini	Cloud	Gemini Pro and other models

Quick Start

Web App

git clone https://github.com/eir-space/eir-open.git
cd eir-open/apps/open-medical-scribe
npm install
npm start
# Open http://localhost:8787

Desktop App

Download the pre-built Electron app:

Light Build (~1 GB) — Clean app, downloads Whisper model on first run
Full Build (~1.9 GB) — Bundled Whisper ONNX model, works offline immediately

Or build from source:

# Light build (no bundled models)
npm run electron:build:light:mac

# Full build (bundled Whisper + LLM)
npm run models:download:whisper
npm run models:download:llm
npm run electron:build:full:mac

CLI Tools

# Transcribe an audio file
node cli/scribe-transcribe.js --input recording.wav

# Generate a note from a transcript
node cli/scribe-note.js --input transcript.txt --style soap

Note Formats

Format	Use Case
SOAP	General clinical encounters — Subjective, Objective, Assessment, Plan
H&P	History & Physical — comprehensive initial evaluations
Progress	Follow-up visits and ongoing care
DAP	Behavioral health — Data, Assessment, Plan
Procedure	Surgical and procedural documentation
Journalanteckning	Swedish clinical format — Kontaktorsak, Anamnes, Status, Bedömning, Åtgärd

Configuration

All settings can be configured via environment variables or the web UI settings page:

# Transcription
TRANSCRIPTION_PROVIDER=whisper-onnx   # or openai, deepgram, google, berget, whisper.cpp
WHISPER_MODEL=KBLab/kb-whisper-large  # HuggingFace model ID

# Note generation
NOTE_PROVIDER=ollama                  # or openai, anthropic, gemini
NOTE_STYLE=soap                       # or hp, progress, dap, procedure, journalanteckning

# API keys (only needed for cloud providers)
OPENAI_API_KEY=sk-...
ANTHROPIC_API_KEY=sk-ant-...
GOOGLE_API_KEY=...

Architecture

open-medical-scribe/
  src/
    providers/
      transcription/    # Pluggable transcription backends
      note/             # Pluggable LLM note generators
    services/
      scribeService.js  # Core pipeline: transcribe -> generate note
      promptBuilder.js  # Note-style-specific LLM prompts
      privacy.js        # PHI redaction utilities
    server/
      createApp.js      # HTTP server and routes
      streamHandler.js  # WebSocket streaming
  public/               # Web UI
  electron/             # Desktop app wrapper
  cli/                  # CLI tools
  test/                 # Tests (Node.js built-in runner)