API Reference
Rerank
Re-rank candidate documents by query relevance.
The Rerank API takes a query and candidate documents, then returns relevance-ranked results. It is commonly used after retrieval in RAG pipelines.
Endpoint
| Method | Path | Description |
|---|---|---|
POST | /v1/rerank | Re-rank candidate documents |
Request Fields
| Field | Type | Required | Description |
|---|---|---|---|
model | string | Yes | Rerank model ID |
query | string | Yes | Query text |
documents | array | Yes | Candidate documents, as strings or objects |
top_n | integer | No | Return the top N results |
return_documents | boolean | No | Include the original document in each result, default false |
max_chunk_per_doc | integer | No | Maximum chunks per document, depending on channel support |
overlap_tokens | integer | No | Chunk overlap token count, depending on channel support |
curl
curl https://api.tensoraxis.ai/v1/rerank \
-H "Content-Type: application/json" \
-H "Authorization: Bearer $TENSORAXIS_API_KEY" \
-d '{
"model": "jina-reranker-v2-base-multilingual",
"query": "What is an API gateway?",
"documents": [
"An API gateway handles authentication, routing and rate limiting.",
"A vector database is used for similarity search.",
"Object storage stores files."
],
"top_n": 2,
"return_documents": true
}'Response Shape
{
"results": [
{
"index": 0,
"relevance_score": 0.98,
"document": "An API gateway handles authentication, routing and rate limiting."
}
],
"usage": {
"total_tokens": 42
}
}When return_documents is false, results usually include only index and relevance_score.