Rerank

Overview

Rerank receives a query and a set of candidate documents and returns them in relevance order. It commonly follows vector, keyword, or hybrid retrieval to select the most relevant items from a limited candidate set.

API access

Base URL

https://api.agtcloud.cn/v1

Rerank documents

The following examples use the Jina-style POST /v1/rerank request.

curl 'https://api.agtcloud.cn/v1/rerank' \
  -H 'Authorization: Bearer your-api-key' \
  -H 'Content-Type: application/json' \
  --data '{
    "model": "MODEL_ID",
    "query": "How should I store an API key?",
    "documents": [
      "Store API keys in server-side environment variables or a secret manager.",
      "The model marketplace lists available models.",
      "Never expose a real API key in browser code or a public repository."
    ],
    "top_n": 2
  }'

Parameter	Required	Description
`model`	No	Rerank model ID; whether it can be omitted depends on the selected format
`query`	Yes	Query used to judge relevance
`documents`	Yes	Candidate documents to reorder
`top_n`	No	Number of results to return

Read the result

Results are ordered by relevance. Use results[].index to map an item back to the source documents, and read its relevance score from results[].relevance_score.

Choose a request format

Format	Request path	API Reference
Jina	`POST /v1/rerank`	Jina Rerank
Cohere	`POST /v1/rerank`	Cohere Rerank
Xinference	`POST /v1/rerank`	Xinference Rerank

Understand relevance scores

A relevance score is not a probability.
Scores are primarily useful for comparing candidates within the same request.
Evaluate score thresholds and top_n with your own application data.