Create embeddings
Returns one embedding vector per input string. The request and success bodies follow the OpenAI embeddings API shape; error bodies use the standard B4M error envelope. model is any embedding model this deployment supports (OpenAI, Voyage AI, Amazon Bedrock, or a self-hosted Ollama model); input is a string or up to 2048 strings, at most 300000 tokens in total and each within the model context window. The input count is further capped so inputs x output width stays within 196608 values (128 inputs at 1536 dimensions) - page larger jobs across requests. Pre-tokenized input (arrays of token ids) is not accepted; clients that send it by default, such as LangChain's Python OpenAIEmbeddings, must be set to send strings (check_embedding_ctx_length=False). dimensions shortens the vector on models that support it (OpenAI text-embedding-3-*, and Voyage models with several published widths). Billed in credits per input token from the per-model embedding rate, reserved before the provider call and refunded if it fails. Authenticate with an API key (b4m_live_) carrying ai:generate, or a JWT.
Request
- Base URL: https://your-deployment.example.com
- URL: https://your-deployment.example.com/api/v1/embeddings
- Auth: one of:
- HTTP bearer
- API key in header x-api-key
Request body
Example request
{
"model": "text-embedding-3-small",
"input": [
"The quick brown fox",
"jumps over the lazy dog"
],
"dimensions": 1024
}Response
One embedding per input, in input order.
Example response
{
"object": "list",
"data": [
{
"object": "embedding",
"index": 0,
"embedding": [
0.0023,
-0.0093,
0.0158
]
},
{
"object": "embedding",
"index": 1,
"embedding": [
-0.0112,
0.0047,
0.0201
]
}
],
"model": "text-embedding-3-small",
"usage": {
"prompt_tokens": 9,
"total_tokens": 9
}
}Changes
No recorded changes to this operation across all 1 revision of this API.