hjLabs AI Playground
Starting the on-device runtime…
Turn sentences into vectors with all-MiniLM-L6-v2 (384 dimensions, 23 MB), bge-small-en-v1.5 or mxbai-embed-large — the same encoders people run server-side for semantic search and RAG, executed here on your GPU. You get the raw vector, cosine similarity between any two inputs, and a copyable JSON payload. For prototyping a retrieval pipeline this removes both the API bill and the round trip: a MiniLM embedding takes a few milliseconds once the model is warm.
text embeddings onlinesentence embeddings browserall-minilm-l6-v2 demofree embedding api alternativebge embeddingscosine similarity toolrag embeddings localvector embeddings generator
Yes, within quantization error. The ONNX export is the same network; at fp32 the outputs match the PyTorch reference to several decimal places, and q8 shifts values slightly without meaningfully changing similarity ranking.
You can embed as many strings as you like — there is no quota — but the browser tab is doing the work, so batch through a few thousand chunks rather than a million. Keep the tab focused; browsers throttle background tabs.
No. The model weights are downloaded from Hugging Face to your browser once, and every inference after that runs on your own GPU through WebGPU. Your text, images and audio are never sent anywhere — hjLabs.in has no inference server and no way to see your input.
Small models do. Recent mobile Chrome and Safari expose WebGPU, but memory is the binding limit — stay under roughly 200 MB of weights on Android, and expect iOS to reclaim the tab sooner. The catalog shows every download size up front so you can pick something that fits.
That depends on the individual model, not on this site. Each model page links to its Hugging Face repository, where the licence is stated — most of the catalog is Apache-2.0 or MIT, a few are research-only. We add nothing on top: hjLabs.in claims no rights over anything you produce here.