Embeddings

Create embeddings with OpenAI, Gemini, Mistral, Fireworks, Together and Bedrock models through one API.

The gateway serves OpenAI’s Embeddings API at /v1/embeddings. The model routes to any provider that embeds, on your own key or Respan’s credits, and the answer comes back in OpenAI’s shape.

Pick a model: filter by output type embeddings on the model catalog , or read Models.

Quickstart

curl https://gateway.respan.ai/v1/embeddings \
  -H "Authorization: Bearer $RESPAN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "text-embedding-3-small",
    "input": ["How do I reset my password?", "Where is my order?"]
  }'

import os
from openai import OpenAI

client = OpenAI(
    base_url="https://gateway.respan.ai/v1",
    api_key=os.environ["RESPAN_API_KEY"],
)

result = client.embeddings.create(
    model="text-embedding-3-small",
    input=["How do I reset my password?", "Where is my order?"],
)
print(len(result.data[0].embedding))

import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://gateway.respan.ai/v1",
  apiKey: process.env.RESPAN_API_KEY,
});

const result = await client.embeddings.create({
  model: "text-embedding-3-small",
  input: ["How do I reset my password?", "Where is my order?"],
});
console.log(result.data[0].embedding.length);

Options

Send the request as JSON. A form-encoded body is refused with 400.

Providers

OpenRouter models are not served for embeddings, because OpenRouter’s vendor on the gateway speaks chat only.

Respan parameters

Add customer_identifier, metadata and the other Respan parameters in the body (or extra_body) or the X-Data-Respan-Params header, as on chat. Tagging embedding calls with the same customer as their chat calls lets you see a customer’s full spend. See Respan parameters.

result = client.embeddings.create(
    model="text-embedding-3-small",
    input="Where is my order?",
    extra_body={"customer_identifier": "user_123"},
)

Embedding requests are not redacted, so an organization with PII redaction on is refused.

Pricing and logs

• Each call is one trace with one embedding row, priced on its input tokens. The row keeps the input texts (input_text or input_texts), not the vectors.
• Before a call on Respan’s credits is sent, the gateway holds credits for the input at the model’s input rate.
• On Respan’s credits, a model without an input price is refused with 400 before anything is held or sent.
• An answer on Respan’s credits that reports no input tokens is refused with 502 no_billable_usage and charged nothing.
• Embeddings pass the same checks as chat (limit policies, RPM, key usage cap, customer budgets), with no TPM limit and no customer rate limit.

Next steps

• Text: routing and billing basics.
• Budgets and limits: cap spend per key or customer.
• Respan parameters: tag calls with a customer and metadata.
• Logs: find each embedding row.