The models midden runs
midden works out what each thing you save is about with small models that run on your device. They ship inside the extension and it can't fetch others, so what you save is never sent to a model somewhere else.
| Job | Model | Licence |
|---|---|---|
| Text embeddings | bge-small-en-v1.5 |
MIT |
| Finding text | PP-OCRv4 detection | Apache-2.0 |
| Reading text | PP-OCRv5 English recognition | Apache-2.0 |
| Speech to text | whisper-tiny.en |
Apache-2.0 |
| Speech detection | Silero VAD v5 | MIT |
| Images by content | SigLIP base patch16-224 | Apache-2.0 |
| QR codes | zxing-cpp, or the browser's BarcodeDetector |
Apache-2.0 |
- English only, for now: the embedder, text reading and speech models are all English.
- The phone runs none of them. It searches what your other devices have already worked out.