Hello Pierre-Gilles, hello everyone. It’s been a few years now that I’ve been using/testing Gladys. I want to clarify that I am not a Gladys Plus subscriber (as a tenant, I can’t really push its use). I used to launch scenes or view camera snapshots from the discussion page or from Telegram. It was very convenient, it allowed me to know the weather in the morning when I woke up, or to know if a family member had come home, or even to follow the course of Bitcoin by launching a scene. In short, very practical and super effective. But here’s the thing, since the last update, instead of the usual response I have a forced message: « The chat uses Gladys Plus AI and is only available with a Gladys Plus subscription. [Subscribe to Gladys Plus] ». I suspect this is not a bug or a configuration error. My question is, why remove a very practical feature and instead « force » a subscription to continue using a function that was once so simple to access? I understand that you need to find a way to make money, and with free software that’s not easy. It is indicated on the website: « Gladys is self-hosted: your home automation data stays on your machine. No mandatory cloud, no tracking. » By using an AI, this is no longer the case. Gladys is a self-hosted and autonomous solution, using an AI cuts all that principle in my opinion. It can be useful in some cases, but it shouldn’t become an automatism, and, I think, it shouldn’t cut access to historical features in favor of all AI. Am I the only one in this situation? In any case, a big thank you for having had this crazy idea one day, to invent Gladys Assistant.
Hello @ZoLTRoN, and welcome to the Gladys community 
Thank you for using Gladys all these years, and especially for taking the time to write this detailed message. This is exactly the kind of feedback that helps me make better decisions for the project.
The feature to chat with Gladys you mentioned has been around since the early days of v4, long before the arrival of modern LLMs. It was based on a relatively simple system, with only a few predefined commands and very little flexibility in phrasing.
Recently, with the arrival of modern AI models like ChatGPT, Claude, or Gemini, I realized that this interface often gave new users a bad first impression. Used to being able to converse naturally with these tools, they expected to find a similar experience in Gladys. As soon as they deviated from the few planned commands, Gladys would respond that she didn’t understand, giving the impression of a limited, or even “broken” assistant.
I also received a lot of feedback from people discovering Gladys (especially users coming from Home Assistant or content creators testing the product) who found this tab useless or disappointing compared to current standards.
On top of that, there was a technical constraint: this feature required loading and training a local model at startup, which consumed resources and extended Gladys’ launch time.
For all these reasons, I decided to remove it in favor of the AI assistant. Honestly, I felt that this old feature was hardly used anymore. Your feedback shows that I was wrong about that 
Regarding the self-hosted aspect, Gladys’ philosophy hasn’t changed: the entire home automation part continues to work locally and without mandatory cloud. The AI features are an additional service, but they don’t replace Gladys’ autonomous operation.
In your case, I see a solution that might meet your needs: the MCP server (Documentation). It allows you to connect Gladys to an AI client of your choice to control your home and retrieve information exactly as with the built-in AI assistant.
You can use ChatGPT, Claude, Mistral, or even a self-hosted solution. For example, I made a video where I used OpenClaw on Telegram, connected to Gladys via MCP. This feature doesn’t require Gladys Plus!
I don’t know if this would fully meet your needs, but if you’re interested, I can help you look at this solution in more detail.
Thank you again for your feedback, and thank you for using Gladys for so long
Hello Pierre-Gilles, and thank you for responding so quickly!
I understand your argument about the difference between conversations with an AI and the few commands originally planned in the chat. I just find it a shame to rush into AI at all costs where, in my opinion but it’s personal, there isn’t really any interest. Let me explain
On Telegram, I only had to write one or two words and I got my answer, for example « camera 1 » and boom, the photo from « camera1 » arrived. With the AI, you have to make a whole sentence, which I find less intuitive « send me a photo of my camera 1 ». Plus, you need a subscription, which is immediately less plug and play!
For my part, I live in the countryside where fiber has not yet made an appearance and where ADSL is execrable. My connection is made with a 4G router, it works well but it’s not reliable, outages are frequent hence the need for a self-hosted solution.
I also hear the solution of using a self-hosted AI. But my Gladys runs on a small server, I already tried to install Ollama there, 5 minutes to respond to a word!!! Installing it on a separate machine has a cost, especially these days (and thanks to AI, by the way!), plus the electricity consumption…
I love Gladys, as I said, I’ve been using it independently for a long time. It’s becoming more and more stable (a few versions back I had to restart Docker from time to time but that hasn’t been the case for a long time), there are new features regularly, it’s really a great solution.
Wouldn’t it be possible to have an option, even a hidden one, to activate the old discussions for those who don’t need an AI? You often talk about making Gladys as plug and play as possible. Connecting an MCP server, paying a subscription to an AI… it complicates things a bit, doesn’t it?
Anyway, thank you again for taking the time to answer me, that’s also what makes Gladys a different solution with a real identity.
Have a good day
Hello @ZoLTRoN, and thank you for taking the time to share your feedback.
I completely understand your desire to bring back this feature. It’s always frustrating when a tool or a usage you’ve grown accustomed to evolves or disappears.
On my end, if I’m hesitant to reintroduce it, it’s mainly for maintenance reasons and the overall complexity of the product. This part of the code hasn’t been maintained for several years and represents about 2,700 lines of code, with a fairly heavy NLP library that adds about fifteen dependencies and which is also no longer well maintained. Bringing this feature back would mean reintroducing “dead” code that will eventually need to be replaced. I’m afraid it would only postpone the problem rather than solve it durably.
I completely understand this constraint, and that’s precisely why Gladys continues to function locally for most of its features.
However, if the goal is to interact with Gladys via Telegram to receive photos, an Internet connection remains indispensable since Telegram itself is a cloud service. In this context, the requests sent to an AI (a few hundred bytes of JSON) are generally much lighter than sending an image from a camera.
I’d be curious to know the configuration of your mini-PC as well as the model you tested. During my tests on a mini-PC (on YouTube here: https://youtu.be/ZhqBSgQCO4o?si=vK0_e6k4TW0T9VA3), with a lightweight model, I got perfectly usable performance. Of course, these models are less powerful than those offered via Gladys Plus, but they can already offer an interesting experience for minimal local use.
I really think this could be the solution for you!
Have you tested the latest, lightest Gemma 4 models? (Released two weeks ago)
The lightest version uses 800 MB of RAM and is designed to run on mobile. You can even do image analysis on the 1.1 GB RAM version 
Hello, and thank you again for responding so quickly, and especially for taking an interest in my request. The server on which Gladys is installed, along with many other things, is a Debian 13 on an HP ProLiant MicroServer Gen 10. https://adeo-informatique.fr/wp-content/uploads/2020/05/Quickspecs_HPE_MicroServer_Gen10.pdf With a dual-core processor @1.6GHz and 8GB of RAM at the time I tested Ollama, it now has 24GB. I don’t remember which model I used, but it was impossible, one word every five minutes… I’ll try again now that there are more models available, especially smaller and more efficient ones. Thank you for the information and have a good end of day.