Embedding models used by Joplin for on-device semantic search.
Models are distributed as .tar.gz archives attached to releases in this repo. Joplin downloads the active model on first use of AI features and caches it under the user's profile directory.
Microsoft's multilingual-e5-small (MIT-licensed), ONNX-quantized by Xenova. 384-dim embeddings, 100+ languages, ~140 MB.