Content Table

Prism Endpoint:

AI Matching


Usage

Description

This endpoint evaluates the relationship between a target text and multiple candidates using vector embeddings, returning an ordered list with precise similarity scores. Perfect for intelligent search, categorization, or recommendation engines.


Method & Path

Use https://prism.optical-labs.ca/text/ai-matching to access this endpoint. The method must be POST and the header should contain your secret API key, as shown below:


Body Fields

Field Name Type Description
target String (required) Text with a length between 1 and 500 characters
candidates Array (required) 1 to 500 strings, each with a length between 1 and 500 characters
strictness Integer (optional) Level of matching strictness from 0 to 10

Valid Response

Each successful request to this endpoint costs 2 to 35 credits, depending on the number of candidates. The price formula is 2 + ( candidates.length / 15 ). Your usage and remaining credits are always returned in the meta object of the response.


Interactive Preview

Because of the high-compute nature of this endpoint, the interactive preview is not available.


Examples

Basic Request

The simplest request you can make, using only the target and candidates fields:

The endpoint reorders the candidates from highest to lowest match. The response also includes the full configuration used for the request, including default values for any undefined parameters:


High Strictness

Using a high strictness can be very practical when searching for near-perfect matches:

The candidates are ordered by a hidden raw match score, so the order always stays the same regardless of the strictness level used, even when multiple confidence scores are 0:


Low Strictness

Using a low strictness is most useful for recommendation engines, where you want to find related concepts rather than exact matches:

The confidence scores are now a lot higher because they are all somewhat related to the target: