Keep your SDK.
Change the base URL.

JavaScript / TypeScript

import OpenAI from "openai";
const client = new OpenAI({
  apiKey: process.env.EMBEDDINGS_API_KEY,
  baseURL: "https://api.embeddings.app/v1",
});
const result = await client.embeddings.create({
  model: "bge-m3",
  input: "The quick brown fox jumps over the lazy dog.",
  encoding_format: "float",
});
console.log(result.data[0].embedding.length); // 1024

Pass encoding_format: "float" explicitly, including with OpenAI SDK v6. Use "base64" for little-endian float32 vectors.

Python

from openai import OpenAI
import os
client = OpenAI(api_key=os.environ["EMBEDDINGS_API_KEY"],
                base_url="https://api.embeddings.app/v1")
result = client.embeddings.create(model="bge-m3",
                                 input="A document to search.",
                                 encoding_format="float")
print(len(result.data[0].embedding)) # 1024

Raw HTTP and batches

curl https://api.embeddings.app/v1/embeddings \
  -H "Authorization: Bearer $EMBEDDINGS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"bge-m3","input":["Workspace pulse","收件箱"],"encoding_format":"float"}'

Document search

Upload up to 60,000 UTF-8 bytes of plain text. The original is private in R2; deterministic chunks use BGE-M3 and a 1,024-dimensional cosine Vectorize index.

curl https://api.embeddings.app/v1/documents \
  -H "Authorization: Bearer $EMBEDDINGS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"name":"guide.txt","text":"Our product helps teams search their documents."}'

A successful ingest returns a mutation ID and index_status: "pending". Vectorize visibility is asynchronous; retry search after the index catches up.

curl https://api.embeddings.app/v1/search \
  -H "Authorization: Bearer $EMBEDDINGS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"query":"How do teams find documents?","top_k":5}'

Search returns scored chunks and source metadata from your workspace namespace. Pass those sources to your preferred answer model. Model execution for generated answers is outside this API.

Endpoints

Errors and spending limits