Ferric embeds text with a real embedding model (Qwen3-Embedding) on your GPU, so you can search by meaning — not keywords — entirely in the tab. This is the retrieval half of RAG: your documents and queries never touch a server. The same pure-Rust code as the cloud, and the embeddings are bit-identical to the native build.
Try it
Edit the corpus if you like — each line is embedded once on your GPU. Then ask in your own words; the closest lines by cosine similarity rank to the top. Note it matches meaning, not words (e.g. “how plants make food” finds photosynthesis).
Why it matters
Your corpus and query are embedded and compared in the tab. No embeddings API, no vector DB in the cloud, no data leaving.
A real 0.6B embedding model on WebGPU — it matches paraphrases and concepts, the foundation of retrieval-augmented generation.
Ferric's native server exposes the OpenAI /v1/embeddings endpoint too — point any RAG stack at it.
The exact pure-Rust model, on the browser's GPU — embeddings match the native build to the bit.