About 1avance

1avance is a free AI assistant built on an open model rather than on one of the big closed AI platforms. We wanted to show that a useful, web-connected assistant can rely on a model that anyone can inspect and run.

Why an open model?

Most well-known chatbots run on closed models. You can talk to them, but only through the company that built them, on its terms, and nobody outside that company can inspect how they work. Open-weight models take a different approach: the trained model is published in full, so anyone can download it, study it and run it on their own machines.

For a small independent project, that changes everything:

• Independence. No single provider can change the price, the rules or the model behind our backs.
• Choice of where it runs. The same model can run on specialist providers or on our own server, and we can switch at any time.
• Transparency. The model we use is public and documented, so its strengths and weaknesses are known.

The model behind the answers

1avance runs Qwen 3.6 35B-A3B, released in 2026 by the Qwen team at Alibaba under the Apache 2.0 license. It is a mixture-of-experts model: it contains about 35 billion parameters in total, but only around 3 billion are active for each word it produces. That design gives it the knowledge of a large model with the speed of a much smaller one, which is what makes it practical on a server without a graphics card.

Like every language model, it learned from data that stops at a certain date. That is why we connected it to the web: when your question is about something recent, it searches first and then answers.

What happens when you send a message

1. Our server adds a few instructions to your message: today's date, the interface language, and anything the assistant has saved about you.

2. The model reads the conversation. If it needs fresh information, it asks for a web search.

3. A metasearch engine running on the same server queries several search engines, and the server reads the most relevant pages.

4. The model writes its answer, citing the sources it used, and the text appears on your screen as it is written.

To keep answers fast, the model usually runs on specialist providers through the OpenRouter service. Our own server runs the very same model and takes over if those providers are unavailable. Either way, the same model answers you.

How we treat your data

We store your conversations and saved notes so that you can find them again and get more relevant answers. We do not train any model on them, and we do not sell them. You can see and delete everything from the My account page. The details are in our privacy policy.

Who we are

1avance is an independent, non-commercial project published by Olivier Muhleisen, who also runs a network of free websites: maps, webmaster tools and practical guides. Ideas, bug reports and questions are welcome at contact@1avance.com.

Curious about what the assistant is good at? Take a look at what 1avance can do or the FAQ.