Vectors for your documents, computed in France.

One call, one vector per text

Send model and input — one string or an array of strings — to POST /embeddings. The answer is a list with one vector per input, in the order you sent them, and a usage block with the input tokens you pay for. The key is the same as for chat (how keys work).

curl https://evanesca.app/api/v1/embeddings \
  -H "Authorization: Bearer $EVANESCA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen3-embedding-8b",
    "input": ["Contract clause 4.2", "Payroll export, March"]
  }'

import os
from openai import OpenAI

client = OpenAI(base_url="https://evanesca.app/api/v1", api_key=os.environ["EVANESCA_API_KEY"])

vectors = client.embeddings.create(
    model="qwen3-embedding-8b",
    input=["Contract clause 4.2", "Payroll export, March"],
)
print(len(vectors.data[0].embedding))  # 4096

import OpenAI from "openai";

const client = new OpenAI({ baseURL: "https://evanesca.app/api/v1", apiKey: process.env.EVANESCA_API_KEY });

const vectors = await client.embeddings.create({
  model: "qwen3-embedding-8b",
  input: ["Contract clause 4.2", "Payroll export, March"],
});
console.log(vectors.data[0].embedding.length); // 4096

package main

import (
	"fmt"
	"io"
	"net/http"
	"os"
	"strings"
)

func main() {
	body := strings.NewReader(`{"model": "qwen3-embedding-8b", "input": ["Contract clause 4.2", "Payroll export, March"]}`)
	req, err := http.NewRequest("POST", "https://evanesca.app/api/v1/embeddings", body)
	if err != nil {
		panic(err)
	}
	req.Header.Set("Authorization", "Bearer "+os.Getenv("EVANESCA_API_KEY"))
	req.Header.Set("Content-Type", "application/json")
	res, err := http.DefaultClient.Do(req)
	if err != nil {
		panic(err)
	}
	defer res.Body.Close()
	out, _ := io.ReadAll(res.Body)
	fmt.Println(res.Status, string(out))
}

{
  "object": "list",
  "model": "qwen3-embedding-8b",
  "data": [
    { "object": "embedding", "index": 0, "embedding": [ … ] },
    { "object": "embedding", "index": 1, "embedding": [ … ] }
  ],
  "usage": { "prompt_tokens": 14, "total_tokens": 14 }
}

The embedding models

All of them run in France, at a French operator with no US parent. Dimensions were measured with a real call; the context window is the limit the operator announces.

Embedding models

Model	ID	Dimensions	Context (tokens)	Licence	Run by	Where	Price / M input tokens	Card
Qwen3 Embedding 8B	qwen3-embedding-8b	4,096	40,960	Apache-2.0	Scaleway	Paris, France	€0.168	Model card
BGE Multilingual Gemma2	bge-multilingual-gemma2	3,584	8,192	Gemma Terms of Use	Scaleway	Paris, France	€0.168	Model card

Prices per million input tokens, margin and VAT included. An embedding has no output price.

How to choose

Decide before you index anything: the vector length becomes part of your storage schema, and switching models later means computing every vector again.

• Shortest vectors: BGE Multilingual Gemma2, 3,584 dimensions — lighter to store and to search.
• Longest vectors: Qwen3 Embedding 8B, 4,096 dimensions.
• Longest texts: Qwen3 Embedding 8B reads up to 40,960 tokens per input, the others stop at 8,192. Cut your documents to fit.
• Lowest price: Qwen3 Embedding 8B and BGE Multilingual Gemma2, €0.168 per million input tokens.
• Licence: Qwen3 Embedding 8B under permissive licences (Apache-2.0); BGE Multilingual Gemma2 comes under Google's Gemma Terms of Use, not an open-source licence — read them before you build on it.

Retrieval and answer, both in Europe

RAG takes two models: one turns your documents and the question into vectors, the other writes the answer from the passages you find. With an embedding model from this page, run in France, and any of Evanesca's chat models, all run in Europe, both halves stay in Europe, with the same key.

1. 01
   Chunk

   Split your documents into passages that fit the model's context window.

2. 02
   Embed and store

   Send the passages to /embeddings and keep the vectors in your vector database, next to the text.

3. 03
   Retrieve

   Embed the question with the same model and look up the nearest passages.

4. 04
   Answer

   Send the question and those passages to /chat/completions, with one of Evanesca's chat models, all run in Europe.

Your vectors live in your own store. Of an embeddings call, Evanesca records the model, the token count and the cost — no text, no vector.

Billing

An embeddings call uses input tokens only, paid at the model's input price from your prepaid credit. No output price, no subscription: credit starts at €2 and never expires (pricing).

Frequently asked questions