Build with Evanesca

An OpenAI-compatible API on AI models run in Europe: the guides, the API reference and the tools that plug into it.

Catalogue checked on 20 Oct 2026

AI modelsrun in Europe.

OpenAI's format: the SDKs you already use.
Every response names its operator and its place.
Evanesca keeps no copy of your prompts.

Evanesca is first a private AI agent and chat, run in Europe; its API gives your own code the same models, in OpenAI's chat-completions format. Point your SDK or HTTP client at the base URL below, send a key in the Authorization header, and your existing code calls 17 models run in Europe — in France by Scaleway and OVHcloud (as a fallback), in Mistral AI's own EU region, in Germany by AKI.IO and in Switzerland by Infomaniak. No call leaves Europe, and every response names the operator that served it and the place.

SoonOpening soon: what follows describes the service as it will work.

Send your first requestA key, the base URL and one call in cURL, Python or JavaScript — then streaming and the response headers.Explore the APIEvery operation: parameters, examples, responses, where it runs and what it costs.Create imagesFLUX.1 [schnell] and PhotoMaker V2, run by Infomaniak in Switzerland and billed per second of generation.Embeddings for searchVectors for your documents, for search and RAG, billed per input token.Choose a modelWhere every model runs, what each operator keeps, and the compliance record GET /models returns.MCP serverConnect Claude Code, Cursor or Cline with a key — ChatGPT and Claude with OAuth 2.1.Use Evanesca in ChatGPT or ClaudeAdd Evanesca as a custom connector, sign in with OAuth 2.1 and approve the rights it asks for.

Base URL

https://evanesca.app/api/v1

Every endpoint lives under this URL. Send your key in the Authorization: Bearer header — keys start with sk-eva- (authentication). The catalogue, GET /models, needs no key (the reference).

Use your existing tools

OpenAI's official SDKs for Python and JavaScript work as soon as you change the base URL: there is no Evanesca library to install (first request). n8n and Microsoft Foundry have step-by-step guides; LangChain, the Vercel AI SDK, Open WebUI and LibreChat are on the integrations page. Claude Code, Cursor and Cline connect over MCP, ChatGPT and Claude through the OAuth connector.

Evanesca serves OpenAI's Chat Completions format and, on the same models, Anthropic's Messages format (POST /messages, for Claude Code and Anthropic's SDKs). It does not serve OpenAI's newer Responses API (/responses): a tool that defaults to it must be switched to chat completions — the integrations page says where.

Documentation for agents

The llms.txt index lists these pages and every API operation. The full text gives, in Markdown, the home page, the models with their checks and prices, the passport, the sources and a summary of the API — base URL, endpoints, errors. Both are generated from the same data as the site.

Frequently asked questions

Can I use OpenAI's official SDKs?

Yes. The API follows OpenAI's chat-completions, embeddings and images formats: set base_url (baseURL in JavaScript) to Evanesca's base URL and api_key to your key. The Python and JavaScript SDKs, and the tools built on them, work with no other change.

Can a call leave Europe?

No. Every model Evanesca serves runs in Europe, and no call leaves Europe: not as a backup, not when servers are busy — if an operator fails, you get its error. A model Evanesca does not serve answers 404 model_not_found.

Why did my answer come back empty?

Most likely a reasoning model called with a small max_tokens: every token went into thinking, and the answer came back with no content and finish_reason “length”. Leave max_tokens out, or allow it about 1,000 tokens or more.

What happens when my credit runs out?

New calls get 402 insufficient_credit as soon as the credit is at zero or below. Calls already running are still charged and can take the balance slightly below zero; the difference comes off your next credit purchase.

Does Evanesca keep my prompts?

No. Prompts and answers stay in memory while the answer is written, then they are gone; of each call, Evanesca records the model, the tokens and the cost. On the operators' side: when requests cause errors or look abusive, Scaleway may keep and read their content for up to 14 days; and Mistral AI, which runs Mistral Large 4 and Mistral Small 4, keeps API inputs and outputs for 30 days to monitor abuse.

Can the API generate images?

Yes: POST /images/generations with FLUX.1 [schnell], from a text prompt, and POST /images/edits with PhotoMaker V2, which makes portraits from 1 to 4 face photos. Both are run by Infomaniak in Switzerland and billed per second of generation, €0.0092 a second, VAT included — about €0.02 for a FLUX.1 [schnell] image.

Ready for the opening.

Keys are created in your Evanesca account. A question about the API? contact@fleeceai.agency

Get an API key Integrations