The therapeutic frame,
kept on your machine.
TheraFrame records, transcribes, and writes DAP/SOAP notes from your Zoom sessions — entirely on your Mac. No cloud round-trips, no BAA scrutiny, no audio leaving the room where the work happened.
- ✓ Runs entirely on your Mac
- ✓ No PHI transmitted, no BAA needed
- ✓ One-time license, not a subscription trap
In therapy, the frame is everything the work happens inside.
The hour. The room. The fee. The confidentiality. The frame is what makes therapy different from a conversation with a friend — and breaches of it harm the work.
TheraFrame extends that frame to the documentation. The audio never leaves your Mac. The transcript never leaves your Mac. The notes never leave your Mac. Nothing crosses the threshold that wasn’t supposed to.
One Mac. Four steps. Zero round trips to anyone’s server.
The whole pipeline runs locally. Network traffic is limited to the initial model downloads.
- 01RECORD
Stereo capture, physically separated.
Your microphone on the left channel, the remote participant on the right — routed via BlackHole virtual audio. No mixing, no compression, no upload.
- 02TRANSCRIBE
whisper.cpp, Metal-accelerated, on your machine.
Each channel is transcribed independently. Channel-split means perfect speaker separation — physics, not ML diarization.
- 03SUMMARIZE
Notes in DAP, SOAP, or your template.
A local Ollama model writes the chart note from your prompt template. Swap the model, swap the template — your machine, your rules.
- 04REMEMBER
A knowledgebase that accumulates per client.
Each client has a structured walled-garden KB — presenting concerns, goals, important people, themes, homework, risk. Updated after every session, gated by fail-loud validation.
Cloud transcription was the only option. Now it isn’t.
Apple Silicon plus open-weights models means a solo practitioner can run the same caliber of transcription and summarization on a Mac Mini that a SaaS would charge you forever for — and keep the data inside the room.
Eight pieces of machinery, one menu-bar app.
Channel-split diarization
Mic-left, remote-right routing means each whisper run sees one speaker. More accurate than any pyannote model. Zero ML dependencies.
Per-client knowledgebase
Eight-section structured KB per client. Presenting concerns, goals, themes, homework, risk. Updated after every session.
Template-driven notes
DAP, SOAP, or your own. Each template declares its own model, temperature, and context window in its header.
Therapist voice file
A separate KB that captures your clinical voice across all sessions. Notes start to sound like you.
Fail-loud defaults
Bad LLM outputs are rejected. The prior KB is preserved. A retry-once-with-stronger-prefix mechanism recovers most regressions automatically.
Pipeline diagnostics
A live progress bar with streamed logs: audio config, mic permission, volumedetect, KB validation, elapsed timings.
Encrypted at rest
All output lives in a macOS sparsebundle. Eject it when you leave the room.
Bring your own model
Ollama on the back end. Drop in any open-weights model the day it ships.
A Mac that’s already on your desk.
- HARDWARE
- Apple Silicon Mac (M1 or later). 16 GB RAM minimum; 24 GB recommended if you want the large-context KB model.
- DISK
- ~10 GB for the whisper + Ollama models. Output lives in an encrypted sparsebundle you control.
- SOFTWARE
- macOS 13 or later. Zoom (any tier). BlackHole 2ch (free, open-source). Ollama. whisper.cpp (brew install).
- SETUP
- A guided Audio MIDI Setup pass on first run, plus mic-permission grant. Done in about ten minutes.
Questions we get.
01 Is this HIPAA compliant?
02 What about Zoom itself?
03 Will it work for in-person sessions?
04 What if the LLM gets it wrong?
05 Can I customize the note format?
06 Will my notes sound like me?
07 Can I export to my EHR?
08 What does it cost?
One machine. One purchase. Yours to keep.
TheraFrame is in closed beta with a single practitioner. Pricing is provisional until the second and third clinicians come on board.
- One macOS machine, one clinician
- All current note templates
- Local model registry — bring your own Ollama models
- Updates while you’re licensed
- Email support, response within two business days