Are you already using Tasmota with AI? Try this multi-device experiment

[norbert-walter]

Tasmota with AI - A possible scenario for the future

I've been experimenting with a question that I think could become increasingly relevant for Tasmota: Is anyone here already using Tasmota devices directly with AI agents, and if so, what are you doing with them?

For my ADD (AI Device Description) project, I integrated an AI-readable device description into a Tasmota-based device concept. ADD gives an AI additional context about the individual device — what it is used for, what it can do, how it can be controlled, and which constraints or rules apply.

To make this easy to explore without requiring physical hardware, I built a Tasmota simulator that behaves like an ADD-enabled device for the commands needed in the experiments. It has become an experimental platform for exploring what happens when AI agents interact directly with Tasmota-style devices.

The latest experiment is a simulated living room with five Tasmota power switches. Technically, the switches are almost identical. Their ADD descriptions, however, give them different real-world roles and context.

The experiment is simple: give the device descriptions to an AI, tell it what you want to achieve, and see whether it can understand the environment, select the appropriate devices and coordinate them without a predefined scene or automation.

Everything is publicly accessible and can be tested with ChatGPT, Claude, Gemini or local models:

https://github.com/norbert-walter/ai-device-description-add/tree/main/test_scenario

This is deliberately an experimental platform rather than a finished automation solution. I want to explore what works, what fails, which information an AI actually needs, and where deterministic safety mechanisms are required.

So I'm particularly interested in two things: Who is already experimenting with Tasmota + AI, and what happens when you try this scenario with your own AI model and your own tasks?

If you try it, please share your results — especially failures, unexpected behavior and ideas for other experiments.