WebAI StudioBack to Demos
Hot or Cold
EmbeddingsGuess the secret word: every guess is embedded on-device and scored by how semantically close it lands.
Why On-Device?
A whole word game with zero backend: scoring is a local cosine similarity, so it is instant, free to run, and works offline.
Interactive Demo
Semantic Embedder: Loading...
A secret word is picked and embedded locally. Guess it by getting semantically closer.
0 guesses
Guesses, Closest First
Start a game to begin guessing.
How It Works
- A secret word is picked and embedded once, on-device.
- Every guess is embedded locally and compared with cosine similarity — no server ever sees your guesses.
- The temperature tells you how semantically close you are: "ocean" is scorching for "submarine", "banana" is freezing.
- Think in meaning, not spelling — related concepts, categories, and contexts score high.
Implementation Code
demo.js
Keep exploring
56 demos, all running on-device.
Moderation Cascade
The production pattern for cheap moderation: embeddings clear the obvious 90% in microseconds, the LLM judges only the borderline.
Semantic Embedder Prompt API
Chat With Your Document
Fully local RAG: retrieve the relevant passages with embeddings, then answer with the Prompt API — citations included.
Semantic Embedder Prompt API
Semantic vs. Keyword Search
Search a help center by meaning, side by side with keyword search — and watch keywords miss what embeddings find.
Semantic Embedder