OpenBarkPre-alpha

Your consult, written up while you finish the exam.

OpenBark records the visit, transcribes it and drafts the medical record with open models on your clinic's own computer. Every line links back to what was said, so you can check it before you sign.

Download See every feature

Free and Apache-2.0. Windows, macOS and Linux, tested on Windows so far. Pre-alpha: use it alongside your usual records.

Pepper, recheck, ear infection 00:50

Transcript

  1. S1So this is Pepper's recheck for the ear infection. How has she been doing?
  2. S2She stopped shaking her head after about four days, and we finished the drops yesterday.
  3. S1Her weight today is 12.6 kilos, the same as last time. Temperature is 38.5.
  4. S1The left ear canal looks clean now. No redness, no discharge, and she isn't painful when I handle it.
  5. S1Keep cleaning both ears once a week, especially after swimming. No more medication for now.
  6. S1Call us if the head shaking comes back. Otherwise we'll see her at her annual exam in March.

Draft note, recheck

History

Exam

Plan

Rest on a line of the note, or tap it, and the sentence it came from lights up.

Quick to draft, careful enough to sign.

Three things decide whether a scribe belongs in a clinic: whether you can check it, whether the data stays put, and whether it knows a herd from a hamster.

Checkable, line by line

A draft is only useful if you can trust it, so every statement shows where it came from.

Weight 12.8 kg, unchanged

12.8 kg isn't in the transcript. Check it before you sign.

  • Every line links to the moment in the recording it came from, and many mismatched links are flagged.
  • Numbers that aren't in the transcript are flagged before you can sign.
  • Optional: a second speech model re-listens to lines with numbers and likely drug names, and flags what it hears differently.

Private by design

Recording, transcription and drafting all happen on the computer in your clinic.

  • Recordings, transcripts and notes are encrypted on disk, with a recovery key you keep.
  • No telemetry, no analytics, no cloud fallback. Every network call it can make is listed in its documentation.
  • Keep visits on a backed-up drive of your choosing.

Built for every species

Dogs and cats, horses, herds, birds and exotics, with headings that fit each one.

  • 20 templates out of the box: new consults, rechecks, phone calls, dental, surgery, rounds and more.
  • Multi-pet visits get a separate note for each animal.
  • Notes in English, French, Spanish, German, Portuguese, Italian or Dutch, and the app in five of them.

What the evaluation shows so far.

Scored on synthetic consults with known facts. Real-clinic validation comes with the beta.

0numbers in evaluated drafts that weren't in the transcript, across every scripted test consultdocs/eval/M1.md (4 consults, every prompt revision); docs/eval/M2.md (9 specialty consults, 3 GP templates, 3 client documents)
91%of facts captured in SOAP notes (40 of 44) at the end of M1, with the optional second recogniser installed, up from 75%docs/eval/M1.md table, Gemma 4 E4B on synthetic consults
91%of facts captured across 9 specialty consults (89 of 98) in the latest measured builddocs/eval/heading-order.md (current build, Gemma 4 E4B)
13.1 sfrom pressing stop to a saved note, on a 9.6-minute consultdocs/perf/M0.md 'Live recording' (RTX 4080, real-time playback, 30 s target)
20built-in templates: 16 for the record and 4 for ownerspackages/templates/ (20 folders)
25consult languages recognised, with notes in 7 and the app in 5docs/languages.md; engine/os-engine/src/note.rs NOTE_LANGUAGES; apps/desktop/src/i18n.ts UI_LANGUAGES
43 sto import an 8.5-minute recording with the NVIDIA speech pack, against 152 s on the processordocs/adr/0014-gpu-speech-recognition.md (RTX 4080)
5network connections the app can make, each started by the userdocs/network.md (R1 to R5)

Open models, on hardware you already own.

No subscription to a model provider and no account. The models download once, then OpenBark works offline. A graphics card makes it much faster. Timings so far come from one desktop with an NVIDIA graphics card; laptop and Mac timings are still to come.

  • Writes the note

    Gemma 4

    Google's open model, run by llama.cpp on your graphics card or processor. E4B for most computers, 12B where there's room.

  • Transcribes

    Parakeet TDT 0.6B v3

    NVIDIA's open speech model, for English and 24 other European languages. Separate open models (Silero VAD, pyannote and WeSpeaker) detect speech and label speakers as you talk.

  • Checks the numbers

    Whisper large-v3-turbo

    OpenAI's open model, an optional download, listens again to lines with numbers or likely drug names, and flags what it hears differently.

Install it like any other app.

No administrator rights needed on Windows or Linux; on a Mac, a standard account may be asked for an administrator password to use the Applications folder. Uninstalling keeps your visits unless you choose otherwise.

Windows

Windows 10 or 11, 64-bit. 16 GB of memory recommended.

macOS

macOS 13 or later, Apple silicon. 16 GB recommended, 24 GB for the larger model.

Linux

Ubuntu, Debian, Fedora and other recent 64-bit (x86-64) distributions, as a .deb, .rpm or AppImage. Not yet tested on Linux.

iPhone

Planned: an iPhone app for recording in the exam room. It has not been built yet.

Made by a veterinarian, for veterinary teams.

OpenBark was created by Dr. Christopher Pinard and ANI.ML Health Inc. as an open-source tool for the profession, built together with Claude, Anthropic's AI.

How it was built

Talk to us.

Questions, feedback, a bug to report, or a clinic that would like to try OpenBark: write to the team at ANI.ML Health.

we.care@animl.health

Buy us a coffee

OpenBark is free and always will be. If it gives you back some evenings, a coffee helps keep it going.

buymeacoffee.com/animlhealth