Local models and templates
Transcription and summaries use separate models. Download a transcription model in Patter; run a local model server for summaries.
Download a transcription model
- Open Settings → Models → Transcription.
- Select a model and choose Download. Wait for download and verification to finish.
- Choose Save changes.
- Open a conversation with saved audio, switch to Transcript, then choose Transcribe recording.
The current bundled catalogue includes:
| Model | Approximate download | Runtime |
|---|---|---|
| Parakeet v3 | 632 MB | Core ML on Apple Silicon |
| Whisper Base | 148 MB | whisper.cpp |
| Whisper Small | 488 MB | whisper.cpp |
Sizes describe model files before filesystem overhead. Downloads come from pinned Hugging Face revisions and are checked against expected file sizes and SHA-256 hashes. Once installed, transcription runs locally without a network connection.
Verify / repair checks an installed model and downloads files that need repair. Cancel download stops the current download; selecting Download again can resume partial files where the server supports it. Model weights are excluded from library backups and can be downloaded again on another Mac.
For an existing whisper.cpp GGML .bin model, select Whisper — choose a file, choose the file and save. Keep that file available; an external model needs relinking after moving to another Mac. Patter does not accept arbitrary Parakeet model folders. New compatible model versions are added through app updates rather than silently replacing your selected model.
Transcripts retain timing and audio-track information. Microphone and computer-audio labels identify tracks; they do not identify individual speakers. A short synthetic transcription has been verified. Check real calls, accents and longer recordings before relying on the result.
Connect a summary model
Start a local OpenAI-compatible server with a model loaded, then open Settings → Models → Summaries.
| Server | Local server address |
|---|---|
| LM Studio | http://127.0.0.1:1234/v1 |
| Ollama | http://127.0.0.1:11434/v1 |
Enter the address under Local server, choose Find models, select or enter a model name, then Save changes. Use the port your server actually runs on and include /v1. Patter accepts loopback addresses such as localhost, 127.0.0.1 and ::1; remote model servers are rejected and there is no cloud fallback.
Open a conversation's Summary tab and choose Create summary. If no models are found, check that the server is running and has a model loaded. If generation fails, check the saved address, model name and the server's status.
Choose a summary template
Under Settings → Models → Templates → Default, choose General meeting, One-to-one, Interview, Brainstorm, or Custom. Expand Edit instructions to change the selected template's guidance; Reset instructions restores its built-in guidance. Save your changes.
Inside a conversation's Summary tab, expand Template to choose an override or add Extra instructions. Leaving the template on Default follows your saved global choice. Each set of instructions is limited to 4,000 characters and applies to the next generated summary.
Regenerating a summary keeps the previous version in history. Generated versions record the model, template and instructions used. Review generated summaries and action items for accuracy before acting on them.