The Voice Web with maplibre-voice

The experiment maplibre-voice lets you control a map in the browser with voice commands. There are not only fixed commands, but you can speak any command you can think of. The LLM behind this application will try to understand the command and transform it into an API query that returns the data.

The result is parsed and displayed on the map as pins. This dynamic capability sets this project apart from similar attempts at controlling a map with voice commands.

This project and video was made as part of the Mistral Hackathon 2026.

See the project GitHub: https://github.com/tderflinger/maplibre-voice

This is the transcript of the video:

Hello, I'm Thomas, a software developer and AI enthusiast.

Today I want to talk about the Voice Web.

Web applications in the browser that can be operated through voice.

With voice-enabled interfaces, users can speak to websites to retrieve

information or execute actions on websites.

Responses can also be delivered as spoken output, enabling conversational

interaction.

The core technology stack is now mature enough to make this practical.

High-quality speech recognition models such as Voxtral from Mistral are available

today.

The remaining challenge is implementation design.

How to integrate voice interaction patterns effectively into web experiences.

Voice-enabled websites not only improve accessibility for users with disabilities,

but benefit everybody by making tasks more enjoyable and efficient.

For Mistral Hackathon 2026, I built an experimental project that enables voice

control of a browser-based map.

Here's a short demo of how it works.

Zoom in.

Zoom out.

Fly Salzburg, Austria.

Museums.

Fly Munich, Germany.

Show me all French restaurants.

In maplibre-voice, there are a number of fixed commands, like zoom in and museums.

But dynamic queries, the query for French restaurants in this example, are what

makes this experiment most interesting in my opinion.

You can voice any command and the Devstral LMM tries to mold it into an API query

that then gets executed.

The results of that query are then shown as pins on a map, like you can see here.

This is just one example of the voice web.

There are literally millions of use cases waiting to be explored by talented

innovators.

I hope you enjoyed this demo and if you have any questions or feedback,

send me a message or write me here on YouTube.

Thank you!