The ambient scribe that keeps the consultation in the room.
ClinicAVT listens to a consultation and drafts the clinical note, a plain-language summary for the patient and the relevant guideline passages, all on the clinician's own Windows laptop. It fine tunes results with the clinician's own guidance documents, without uploading to the cloud. No audio or text is sent anywhere.
· Runs entirely offline · Uses OpenVino and Intel AI PC processors ·
Demonstrator published by MotionInput Games Ltd, for personal and experimental use only.
It is not licensed for enterprise or clinical use - contact us for licensing information.
How it works
You record and review. ClinicAVT does the rest.
Record
Press Record, or import a file.
Transcribe
Speech to text, labelled Doctor or Patient.
Draft
Note and patient summary.
Guidelines
Matching passages from your own guideline PDFs.
Review
Edit, translate, then copy or export.
Features
Private by design
No audio, text or usage data leaves the laptop. Consultations are encrypted at rest, and audio is discarded after processing.
Speaker-labelled transcripts
Whisper speech recognition with clinician and patient attribution, plus optional voice enrolment.
Clinical notes
Prose or SOAP notes at two lengths, from a choice of three local language models sized to the hardware.
Patient information in 24 languages
A plain-English summary at a UK reading age of about 9, which can be translated into 24 languages, including right-to-left scripts such as Urdu.
Guideline references
Add your own guideline PDFs. The passages that match the note appear beside it, each with a page number and a citation.
Appraisal reflections
Turn a consultation into an anonymised case study and reflection, then export them for your appraisal.
Backup and restore
Password-protected backups of consultations and reflections. Audio is never included.
Works offline
After installation it needs no network connection, so it works on home visits and in rooms with poor signal.
Swappable models
Newer Qwen models can be added without changing any code.
Screenshots
All consultations shown are simulated. Select an image to enlarge it.
Results
Measured on 57 simulated GP consultations, using a Dell XPS 16 (Intel Core Ultra Series 2) and a Dell XPS 14 (Intel Core Ultra Series 3).
Technology
ClinicAVT comprises a native C++20 inference engine built on OpenVINO GenAI and a WinUI 3 desktop client. All models listed below execute on-device across the CPU, integrated Arc GPU and NPU, from 16 GB of system memory with the smallest note model. Where the clinician chooses to save consultations, they are stored in an encrypted SQLite database.
| Task | Model | Precision |
|---|---|---|
| Speech recognition | Whisper Large v3 Turbo | INT8 |
| Voice activity | Silero VAD | – |
| Speaker segmentation | pyannote segmentation 3.0 | – |
| Speaker identification | ERes2NetV2 | INT8 |
| Note writing | Qwen3.5 4B / 9B, Qwen3.6 35B-A3B | INT4 |
| Translation | NLLB-200 600M | INT8 |
| Guideline search | GTE-large | INT8 |
Get started
Windows 11 · Intel Core Ultra Series 2 or later, with Arc graphics
| Note model | Memory | Disk |
|---|---|---|
| Qwen3.5 4B (Available on the Microsoft Store) | 16 GB | 6 GB |
| Qwen3.5 9B (Contact MotionInput Games for review) | 24 GB | 14 GB |
| Qwen3.6 35B-A3B(Contact MotionInput Games for review) | More than 32 GB | 24 GB |
The first launch prepares the models for your hardware, which takes a few minutes.