An independent Sovereign Website demonstration. Not affiliated with or endorsed by Mistral AI.

Demo 1

Same brief, three brains

Write one brief. This page sends it, through one small function, to a European model, to any model on OpenRouter, and to a model on your own computer, and puts the answers side by side. The code that calls each one is the same; only the address, the key and the model change.

1. Your keys

Your keys stay in this tab. Each goes only to the provider it belongs to, and only when you press Send. This page has no server and saves nothing.

2. The brief

Start from one of these, or write your own

A short limit keeps it quick and cheap. Free tiers have low limits.

3. The brains

Tick the ones to ask. You can type any model id; the lists are suggestions.

4. Send

Nothing is sent until you press Send. Then your browser sends the brief to each provider you ticked, exactly as shown under "What each call sends, and to whom" below, and nowhere else. Nothing goes to this site.

What each call sends, and to whom

This is built from the brief and the brains above, before anything is sent. Your key is shown only by its last four characters.

Calls made from this tab

TimeToModelStatusTime taken (ms)TokensBytes sent

Kept in memory in this tab only, and gone when you reload.

The code, all of it

This is the function every call above goes through, read from the running page, not a copy:

And the only things that differ between the brains:

OpenRouter and local servers such as Ollama and llama.cpp offer the OpenAI-style /v1/chat/completions. Mistral's docs describe /v1/chat/completions as its native format, and name https://api.mistral.ai/openai/v1 as its OpenAI-compatible endpoint, in public preview (chat completion, as read on 20 November 2026). For a request as simple as this one, both returned the same OpenAI-style answer, with choices and usage, in our test on 20 November 2026 (model ministral-3b-2512), so this page uses the generally available /v1.

Can a web page call them?

A browser only sends a request with a key to another site if that site says, in answer to a preflight check, that it allows it (CORS). We checked on 20 November 2026:

EndpointPreflight answerFrom a real browser
api.mistral.aiAllows any page (Access-Control-Allow-Origin: *), Authorization and Content-Type allowedA call from a page in Chromium came back readable: status and JSON body, with usage
api.eu.mistral.aiAllows any page, Authorization and Content-Type allowedA call with a dummy key came back as a readable 401, "Invalid API Key"
openrouter.aiAllows any page, Authorization and Content-Type allowedA call with a dummy key came back as a readable 401
localhostUp to your server. Ollama allows only 127.0.0.1 and 0.0.0.0 pages by default; set OLLAMA_ORIGINS (Ollama FAQ, as read on 20 November 2026)No model server on our test machine, so the browser reported "Failed to fetch"; the page explains what to check

A provider can change its headers at any time; if a call fails with a network error, this page says so plainly.