OpenAI: Embeddings

Generate embeddings for a buffer of claims via OpenAI's embeddings API. Each input claim is vectorized and the response keeps the original identifier and payload, with the produced vector filled in.

For a provider-agnostic embeddings node operating on JSONL files see AI::AIEmbeddings.

Pre-requisite: Install an OpenAI application from Profile > {Organization} > Applications.

Parameters

Application—REQUIRED

Configured OpenAI application.

Input

JSON Embeddings—REQUIRED

Embeddings request like:

Example

{
   "embeddings": [
     {
       "identifier": <String>, # Unique identifier for the claim
       "claim": <String>,      # Text from which the vector will be generated
       "vector": [],           # Empty before embeddings
       "payload": <Json>       # Additional metadata to attach to the claim
     }
   ]
}

Output

JSON Embeddings

Same payload, with each vector field populated:

Example

{
   "embeddings": [
     {
       "identifier": <String>,
       "claim": <String>,
       "vector": [f32, f32, ...],
       "payload": <Json>
     }
   ]
}