Embeddings
Turn text into vectors for semantic search and RAG, previewed directly in the Playground.
Embeddings mode in the Playground turns text into a dense vector using the catalog's embedding model — useful for building semantic search, clustering, or retrieval-augmented generation, and to sanity-check what a piece of text embeds to before you wire it into a larger pipeline.
How it works
Pick the embeddings model from the catalog, paste in one or more pieces of text, and send. Each piece of text gets converted into its own vector.
Playground fields
| Parameter | Type | Description |
|---|---|---|
Modelrequired | select | The embeddings model from the catalog. |
Text inputrequired | text | One string, or multiple lines, to embed in a single request. |
What you see back
The console shows a preview of the resulting vector for each piece of text — its length and the first several values — so you can confirm the embedding was produced as expected. For production use, these vectors are what you'd store in a database or vector index to compare by similarity.
What a good embedding use case looks like
- Semantic search over documents — embed a set of docs once, embed a search query the same way, then compare vectors to find the most relevant matches, even when the query doesn't share exact words with the document.
- Deduplication — embed a batch of records (support tickets, product listings, notes) and compare vectors to find near-duplicates that plain text matching would miss.
- Clustering — embed a large set of text and group it by vector similarity to surface natural categories, without labeling anything by hand first.
Trying a handful of representative texts in the Playground first — a couple of documents you'd search over, or a few records you suspect are duplicates — is a fast way to confirm the embedding model behaves the way you expect before you embed an entire dataset.