Search a vector store using a natural language query.
Authorization
AuthorizationRequiredBearer <token>
In: header
Request Body
application/jsonRequiredqueryRequiredstring | array<string>
The search query in natural language. Accepts either a single string or an array of strings.
Note: Multi-query is currently a request-shape stub for OpenAI compatibility. When an array is supplied, only the first element is used to generate the search embedding; subsequent elements are ignored. The service emits a server-side warning when this happens. Pass a single string for predictable results until proper multi-query semantics are released.
max_num_resultsinteger
Maximum number of results to return (1-20).
5Minimum: 1Maximum: 20filtersobject
Filter criteria for search results.
ranking_optionsobject
Accepted but uses the system's default ranking.
rewrite_queryboolean
Accepted but uses the system's default behavior.
Path Parameters
vector_store_idRequiredstring
The ID of the vector store.
Example
Request
Response
Multi-query request
The query field accepts an array of strings for OpenAI request-shape compatibility, but only the first element is currently used to generate the search embedding. Subsequent elements are ignored and a server-side warning is logged. Pass a single string for predictable results until proper multi-query semantics are released.
The request above behaves identically to passing only "How do I configure the system?" β the second query is dropped.
Cancel file batch
Cancel a file batch that is still indexing; completed files are kept. Covers the path parameters, the cancelled file batch object returned, and an example request and response.
Create response
Generate an LLM answer grounded in your vector stores with the file_search tool (RAG). Covers the request body (model, input, tools, instructions, stream), the response object (output with file_search_call and message items), non-streaming and streaming examples, and the SSE event types.