OpenScribe
AI medical scribe with local Whisper and Ollama processing on desktop, or self-hosted web use with Claude. Free and open source under MIT.
OpenScribe is a free, MIT-licensed AI medical scribe for clinicians who want control over where patient encounters are processed. It records visits, transcribes speech, and produces structured draft notes. Its fully local desktop workflow builds on StenoAI and uses Whisper for transcription and Ollama for note generation.
The desktop app runs on macOS, Windows, and Linux, with local models including Llama 3.2 and Gemma 3. This workflow uses no cloud inference. The recommended minimum is 8GB of RAM and 20GB of free disk space; initial setup includes model downloads.
The self-hosted web app also runs through Docker, but its default workflow uses cloud note generation. Audio stays local for Whisper transcription, while transcript text goes to Anthropic Claude. An alternative sends audio to the OpenAI Whisper API as well. These cloud workflows require API keys and an internet connection.
Notes can follow SOAP or History and Physical formats, or a custom Markdown template suited to a specialty. Whisper supports multilingual transcription when using a multilingual model. A separate local MedGemma workflow generates notes from existing transcripts; it doesn't transcribe audio.
Encounter records stay in encrypted browser storage by default. The app processes audio in memory without saving it and has no analytics, telemetry, or cloud sync. Generated notes require clinical review, and OpenScribe is not HIPAA compliant.
