Back to Home
Embeddings & Search
Embeddings
An embedding is a vector that captures meaning. Similar words end up as nearby points, and that's what makes semantic search possible: instead of matching exact words, AI finds ideas that are close in meaning.
Known words use preset semantic positions that group similar meanings together. Unknown words get a character-based estimate.
Interactive Playground
Statistics
5
Texts
Selected
Closest
Top Score
How It Works
📝
Text
→
🔤
Tokens
→
📍
Embedding
→
📐
Similarity
2D Map (D1 = X · D2 = Y)
Similarity ranking
🎓
Tips
3 tips
Same model = comparable vectors. Embeddings from different models can't be compared directly. Always embed queries and stored items with the same model.
"cat" and "kitten" will have very similar embeddings, semantic meaning is preserved, not just spelling.
Embeddings are fixed once generated. Only re-embed if you change your embedding model.
💡
Key Takeaway
Similar items can be shown as points that are closer together on a map.