Embeddings & RAG

Gonka Broker serves a text embedding model through the standard OpenAI-compatible POST /v1/embeddings endpoint, with the same base URL and API key you use for chat models.

Embeddings turn text into numeric vectors that capture meaning. Store them in a vector database and you can search documents by semantic similarity: the retrieval step behind RAG (retrieval-augmented generation), semantic search, clustering, and recommendations.

The embeddings model

Spec	Value
Model ID	`BAAI/bge-m3`
Vector dimensions	1024 (fixed)
Max input	8,192 tokens per input
Languages	100+, with cross-lingual retrieval

BGE-M3 is BAAI’s open-source multilingual embedding model. Vectors are normalized, so cosine similarity works out of the box. See Supported Models for pricing.

Generate embeddings

Replace YOUR_API_KEY with your actual key (starts with gnk-prx-). If you don’t have one yet, see Create an API Key.

curl https://proxy.gonkabroker.com/v1/embeddings \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "BAAI/bge-m3",
    "input": "Text to embed"
  }'

from openai import OpenAI

client = OpenAI(
    api_key="YOUR_API_KEY",
    base_url="https://proxy.gonkabroker.com/v1",
)

resp = client.embeddings.create(
    model="BAAI/bge-m3",
    input="Text to embed",
)

print(resp.data[0].embedding)

import OpenAI from "openai";

const client = new OpenAI({
  apiKey: "YOUR_API_KEY",
  baseURL: "https://proxy.gonkabroker.com/v1",
});

const resp = await client.embeddings.create({
  model: "BAAI/bge-m3",
  input: "Text to embed",
});

console.log(resp.data[0].embedding);

The response follows the OpenAI embeddings format: a data array with one embedding vector per input, and a usage object with token counts.

Batch multiple texts

Pass an array of strings to embed several texts in one request; one vector comes back per input, in the same order:

{
  "model": "BAAI/bge-m3",
  "input": ["First document chunk", "Second document chunk", "A search query"]
}

Each input can be up to 8,192 tokens; longer inputs return an error instead of being silently truncated, so chunk your documents before embedding. The request body is limited to 2 MiB.

Parameters and limits

Parameter	Supported
`model`	Yes; must be an embedding model
`input`	Yes; a string or a non-empty array of strings
`encoding_format`	Yes; `"float"` (default) or `"base64"`
`dimensions`	No; vectors are always 1024-dimensional

See OpenAI API Compatibility for the full endpoint reference.

Billing

Embeddings are billed on input tokens only; there are no output tokens. Usage is reported in the response’s usage.prompt_tokens, and your per-token rate is locked in when you top up, like every model on Gonka Broker. See Billing & Deposits.

Use with RAG frameworks

Any tool that accepts an OpenAI-compatible embeddings provider works with Gonka Broker: point it at https://proxy.gonkabroker.com/v1, use your API key, and set BAAI/bge-m3 as the embedding model. Connection guides for popular stacks: