---
title: "Document reranking"
description: "Rank candidate documents against a query with the Rerank API."
lastUpdated: 2026-10-22
---

Send `POST https://infistar.ai/v1/rerank` with `Authorization: Bearer YOUR_API_KEY` and `Content-Type: application/json`.

| Field | Required | Meaning |
| --- | --- | --- |
| `model` | Yes | An available reranking model ID |
| `query` | Yes | The user's search text or question |
| `documents` | Yes | Candidate text passages, in their original order |
| `top_n` | No | Number of highest-ranked results to return |
| `return_documents` | No | Include document text when true; default false |

Use the returned `index` to map each result to the original candidate. `relevance_score` expresses relevance for that model; it is not a universal threshold across models.

Reranking follows candidate retrieval; it does not store your documents or replace a vector database. See [RAG integration](/en/integration-guides/rag-retrieval-quickstart).