Searching

A search request is a hybrid of lexical-style query parameters and embedding similarity. You can search by text, by image, or by both at the same time. Add a recommend block to bias results toward items the user interacted with.

Query modes

Text: pass a query string. The database encodes it with the configured embedding model and returns the top_k nearest neighbours.
Image: pass an image_url string. The database encodes the image with the multimodal model and returns the top_k nearest neighbours.
Multimodal: pass both a query and an image_url. The platform fuses the two embeddings before ranking.

Request payload

search-payload.json

1{
2  "query": "How do I deploy a model on QDivZero?",
3  "top_k": 5,
4  "recommend": { "token": "<view_token>", "bias": 0.4 }
5}

The recommend block biases results toward items the user interacted with; see Recommendations for how the token is obtained and how bias is interpreted.

Response shape

The response is an array of hits. Each hit carries the document id, a score (higher is closer in the embedding space), the original content, and the metadata you provided at index time. The same payload is used for the multimodal image_url search when the backing model supports it.