Using AI tools with Domoticz, anyone given it a go?

[waltervl]

I would suggest you ask AI for that....
The answer I got was:

That’s a great question, Walter! Integrating AI with **Domoticz** can definitely open up some powerful possibilities, especially for:

### Surveillance Analysis
Using AI models (like YOLO, OpenCV, or even cloud-based services) to:
- Detect people, vehicles, or unusual activity in camera feeds.
- Trigger Domoticz events (e.g., turn on lights, send alerts) based on detection.
- Filter out false positives (e.g., pets, shadows) to reduce noise.

### Automated Responses
AI can help:
- Adjust heating, lighting, or blinds based on learned patterns or weather forecasts.
- Respond to voice commands more intelligently (e.g., via Home Assistant + Domoticz bridge).
- Predict and preemptively act on sensor data (e.g., water leak detection triggering shutoff valves).

### Smart Reporting
You could use AI to:
- Summarize sensor logs into daily/weekly reports.
- Detect anomalies in energy usage or temperature trends.
- Generate natural language summaries (e.g., “The living room was unoccupied for 80% of the day”).

---

### How to Get Started
If you're curious to experiment, here are a few entry points:
- **Node-RED**: Acts as a bridge between Domoticz and AI services (like OpenAI, TensorFlow, or Hugging Face).
- **Python scripts**: Domoticz supports scripting, so you can run AI models locally or via API.
- **MQTT**: Use it to send sensor data to an external AI service and receive actionable insights.

---

I see no use case for myself except for "Detect anomalies in energy usage or temperature trends."

[HvdW]

I'm using AI for all my dzVents solutions nowadays.
Even more, nowadays when I find an API I ask AI if it can write a script to display the data.
Lately I display data in text sensors.

Here is an example for data from meteoserver.nl which is queried every 2 hours (free)

Screenshot_20250616_235534_Firefox.jpg (358.64 KiB) Viewed 2346 times

It calculates the expected sun radiation using a formula that accounts for low, medium and high cloud coverage.
All made up by AI.

Is it necessary? No.
Is it fun? Yes
From the first setup to completion like I wanted it took a bit more than 1 hour.

[waltervl]

Lately I display data in text sensors.

This will let your database grow rapidly!! check the Light/Switches log settings to minimize this (default 30 days)

[HvdW]

Lately I display data in text sensors.

This will let your database grow rapidly!! check the Light/Switches log settings to minimize this (default 30 days)

Thanks Walter.
Tbh I didn't realize that.
Set it to 10 days.
Tomorrow I'll see the size difference of my domoticz.db

[FireWizard]

Hi,

A couple of years ago I used Tensorflow in combination with intrusion detection.
This was a Node-RED application with a MQTT link to Domoticz.

Domoticz presented the "Alerts", if a person on camera had been detected in a specific area and was responsible for
activation/deactivation of specific (camera) zones.
See a brief description: viewtopic.php?p=268132&hilit=intrusion#p268132

This can also be used, if you want to use "Baby Monitoring" or similar things.

In general, AI combined with camera's, is quite common nowadays.

I abandoned the project, because I used it in a Node-RED instance, installed in a FreeBSD (TrueNAS) jail.
Unfortunately newer versions of Tensorflow did not support FreeBSD.

Regards

[Varazir]

I have created few dzVents script using Gemini, some times you need to wrestle with the AI to get to work

[Derik]

I'm using AI for all my dzVents solutions nowadays.
Even more, nowadays when I find an API I ask AI if it can write a script to display the data.
Lately I display data in text sensors.

Here is an example for data from meteoserver.nl which is queried every 2 hours (free)
Screenshot_20250616_235534_Firefox.jpg
It calculates the expected sun radiation using a formula that accounts for low, medium and high cloud coverage.
All made up by AI.

Is it necessary? No.
Is it fun? Yes
From the first setup to completion like I wanted it took a bit more than 1 hour.

Nice to see.
NOWADAY? Erg benieuwd naar je script.. Dit zou ook met regen kunnen?

[HvdW]

I write English for the sake of this forum.
Regen ook.

Screenshot_20250224_155141_Firefox.jpg (108.75 KiB) Viewed 2232 times

This is for one hour.
Just what I need to know if it is safe to go out shopping without a coat.

https://gpsgadget.buienradar.nl/data/ra ... xx&lon=yyy
Written in dzVents, author DeepSeek
Time to get this was less than an hour.

[Derik]

sorry i now English.. only my dutch is better

[gizmocuz]

I don't think this was the original question.

There are (lightweight) c++ libraries that we can use to train our own model, and feed it with data (temp/power/solar/wind/lux etc etc)
And with this we could make predictions
When you have a system that has been running for some time (like months/years), you could initialy train the model with this data as well

You can also train a model for a light. When someone turns the light on manually always around 6.30am on Monday/Friday, this could be predicted and done automatically

If you known at night your house used 0.7kWh, you could make sure that a battery keeps a minimum load of xx %

and so on...

I tried some coding against 'mlpack', but had issues getting this to work on my main dev machine (cross platform)
I do think this could be a great library to use

[kiddigital]

I see 2 ways for Domoticz where AI might be relevant.

First, using ‘traditional Machine Learning’ to make predictions for sensors. One could train an ML-model based on the historical data of a sensor and with such a model predict a few values in the future for such a sensor. Might be interesting for (light)switches to automatically turn things on or off without the need of a schedule. Just based on historical behavior. For more difficult predictions like weather, the small set of local measurements is insufficient to create a model that has some accuracy. Better use the predictions from already available hardware in Domoticz like Buienrader or OpenweatherMap which offer predictions already.

Second one is making Domoticz ‘accessible ‘ via an LLM like Claude, ChatGPT, Gemini, etc. This way a Chat/Agent gets understanding of the Domoticz setup by providing relevant context to the Chat/Agent. And provide methods for the Chat/Agent to have Domoticz perform actions.
(This second option is one I am currently trying to develop within Domoticz using the Model Context Protocol. Hope to release a Beta in the coming weeks).

Question: what would be actions that one would like to have performed in Domoticz by a (chat)Agent? (Next to the obvious control of switches, sensor setters, reading current sensor values, etc.)

[solarboy]

Live log analysis would be good, for spotting errors and anomalies or whatever you want and then sending "intelligent" notifications. This could also include syslog, and other logs (zwave,zigbee etc).

There are so many things ML could do although I am not sure I would let it have too much control without verification.

[kiddigital]

As mentioned a few weeks ago, I created something as a 'domoticz agent' using MCP. A Draft PR is available now

Curious to hear/read/see if/how domoticz users leverage this and what is missing/wrong/need improvements.

As it is a separate PR and not part of the Beta builds, you either need to build/compile it yourself or download the Linux artifact from a recent GitHub action.

Let me know what the experiences are!

[Thuis]

I used ai to make a python plugin for my google speakers, so Domoticz can "speak" notifications. It creates a mp3 and then talks. Very long sentences is possible. It is added as an notification subsytem. Works great !

I used it to get as close to a "feeling" temperature inside the house. It uses the wind, solar, humidity, temperature, inside and outside, know east and west for the heating of the bricks of the house and so on. Then it steers my AC. Heating as well as cooling. It is always the same temperature in house and that feels great !

I also used ai to make a dzvents to steer the ventialtion of the house, it looks at humidity and CO2 in different rooms. Also it is coupled to the AC script, they work together. So if it is hot inside but cold outside, the ventilstion will work harder to get cool air in from outside so ac does not have to work hard. Other way around, when cold inside and hot outside, it can do the same.

So yes, i use AI tools with Domoticz

[solarboy]

I used it to create a python script I call my "moon-o-meter". it calculates moon brightness by using elevation, cloud cover and moon phase and updates a percentage device. I use that to inhibit my outdoor lighting when the moon is bright because it's beautiful outside with just the bright moon.

[kiddigital]

Live log analysis would be good, for spotting errors and anomalies or whatever you want and then sending "intelligent" notifications. This could also include syslog, and other logs (zwave,zigbee etc).

There are so many things ML could do although I am not sure I would let it have too much control without verification.

Well @solarboy , retrieving domoticz logging is now supported in the (draft) PR for MCP support. So it can be analyzed by an LLM. Results are nice and interesting. It (testing using Claude and VSCode/Copilot) gives good summaries and highlights. Give it a try...

[solarboy]

This is great news. Is it able to be used with offline LLM's ?

[kiddigital]

This is great news. Is it able to be used with offline LLM's ?

It is not specifically designed for that purpose, but as it uses generic MCP (Model Context Protocol) it definitely can be used with offline LLM’s.
Tools like LMstudio, Tome, LibreChat should work fine.
Give it a go and any feedback is welcome.

[waltervl]

As mentioned a few weeks ago, I created something as a 'domoticz agent' using MCP. A Draft PR is available now

Curious to hear/read/see if/how domoticz users leverage this and what is missing/wrong/need improvements.

As it is a separate PR and not part of the Beta builds, you either need to build/compile it yourself ....

Let me know what the experiences are!

That Draft PR is now merged into latest beta 2025.2 so can be used without extra actions.

[kiddigital]

As mentioned a few weeks ago, I created something as a 'domoticz agent' using MCP. A Draft PR is available now

Curious to hear/read/see if/how domoticz users leverage this and what is missing/wrong/need improvements.

As it is a separate PR and not part of the Beta builds, you either need to build/compile it yourself ....

Let me know what the experiences are!

That Draft PR is now merged into latest beta 2025.2 so can be used without extra actions.

Correct, but don't forget it needs to be enabled by specifying

Code: Select all

-mcp

as a startup flag. The log should tell you that it is enabled.