ShareAI Platform Docs

Rerank documents

Provide a query and 1–1000 documents.

View as Markdown
On this page
POST https://api.shareai.now/api/v1/rerank

Rerank documents

Base URL
https://api.shareai.now
Authentication
Bearer API key or approved OAuth access token
Required scope
inference or monetization_gateway

Provide a query and 1–1000 documents. Each document is text, a {"text":"..."} object or a model-supported {"image":"https://..."} object. top_n must be at least 1 and no greater than the document count. Match results to the original document indices.

Before you start#

Set SHAREAI_API_KEY on your server and replace the model placeholder with a currently enabled model. Never expose the API key in browser code. The model, account, credential and provider must all permit this operation. Optional settings below require model support; the API safety bounds are not a promise that every model supports every value.

Send a request#

cURL

curl --fail-with-body --request POST "https://api.shareai.now/api/v1/rerank" \
  -H "Authorization: Bearer $SHAREAI_API_KEY" \
  -H "Content-Type: application/json" \
  --data '{
  "model": "YOUR_ENABLED_RERANK_MODEL",
  "query": "A cat",
  "documents": [
    {
      "text": "A cat on a windowsill."
    },
    {
      "text": "A map of Berlin."
    }
  ],
  "top_n": 1
}'

Request fields#

FieldTypeMeaning and limits
model requiredstringExact enabled public model identifier. Must declare this operation and be allowed by your credential.
query requiredstringNonempty relevance query.
documents requiredarray1–1000 nonempty text strings or single-key text/image objects. Image values must be HTTPS URLs or supported image data URLs, with model image-input support.
top_nintegerOptional result count from 1 to the number of documents. Returned indices refer to the original document array.
return_documentsbooleanOptional provider-supported inclusion of original documents in the result.
providerobjectOptional provider routing preferences. Use provider.only for accepted provider tags, not a prefix in the model ID. Account and credential restrictions still apply.

Read the result#

Success returns HTTP 200 with JSON. Provide a query and 1–1000 documents. Each document is text, a {"text":"..."} object or a model-supported {"image":"https://..."} object. top_n must be at least 1 and no greater than the document count. Match results to the original document indices.

Billing units#

Rerank rates can use input tokens, scored documents, requests or a provider-reported search quantity. No universal search-to-token conversion is assumed. A provider that cannot report a priced dimension is unavailable for that tariff.

Choose the exact enabled model identifier from Console. Its declared operation and input/output capabilities must match this endpoint; a text chat model cannot be substituted for another operation. Your credential, provider and account restrictions still apply.

Pricing and usage#

Rates are published in EUR AI credits for the selected model, provider, units and variants. Only the dimensions in its active tariff are charged. An estimate reserves credits before execution; final charges use measured usage. Prices can depend on resolution, output format or a usage band. Check Console Usage for the final ShareAI debit. Upstream cost fields and media bytes are not converted into invented token counts.

Access and errors#

Use inference for approved network-cost inference or monetization_gateway for an approved Builder markup and accepted billing terms. Identity-only access does not authorize spending. Validation, model authorization, insufficient credits, rate limits and execution failures return a structured error. After a timeout or uncertain submission, inspect Usage or the video job before creating another potentially paid request.

Last updated November 3, 2026

ShareAI Platform Docs
All documentation

Search by title or content.

Ask about this page

Choose an assistant to explore this page. You can also copy the page and paste it into your conversation.

Ask ChatGPTAsk ClaudeAsk GrokAsk ShareAI