Embeddings

The OpenAI embeddings API for text or token-ID input, single or batched.

Create embeddings

model must be an embedding model available to your key (check GET /v1/models); text-embedding-3-small, text-embedding-3-large and text-embedding-ada-002 are the OpenAI targets. Match results to inputs by data[].index. /api/v1/embeddings relays to the same route.

POST/v1/embeddings
curl https://api.experientiallabs.ai/v1/embeddings \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "text-embedding-3-small", "input": ["first document", "second document"]}'

Request fields

Unknown fields are rejected, not ignored.

FieldContract
modelRequired; an embedding model (a chat model is 400 unsupported_capability)
inputRequired. A nonempty string, a list of nonempty strings, a token-ID array (one vector), or a list of token-ID arrays (a batch). No mixing text and tokens, no empty items, negative ids, floats or booleans
dimensionsOptional; only on models that support it (text-embedding-3-* yes, ada-002 no)
encoding_formatfloat (raw HTTP default) or base64. The OpenAI Python SDK asks for base64 and decodes it for you
streamOmit or false; true is unsupported
userOptional, at most 1,024 characters, not forwarded upstream

Token ids must match the model's tokenizer; they are forwarded as is.

import tiktoken

enc = tiktoken.encoding_for_model("text-embedding-3-small")
batches = [enc.encode(t) for t in ["A short document.", "A second document."]]
r = client.embeddings.create(model="text-embedding-3-small", input=batches)

Billing, retries, caching

Billed on provider-reported input tokens (usage.prompt_tokens equals total_tokens); no output tokens. On BYOK your provider bills you. There is no response cache and Idempotency-Key is not honored: a repeated request can be charged again, so store vectors yourself and retry deliberately (the examples set SDK retries to 0). Errors: 400 invalid_parameter (fix param), 400 unsupported_parameter (remove the field).

from langchain_openai import OpenAIEmbeddings

emb = OpenAIEmbeddings(
    model="text-embedding-3-small",
    base_url="https://api.experientiallabs.ai/v1",
    api_key="YOUR_API_KEY",
)
vectors = emb.embed_documents(["A small example.", "A second document."])