That sounds great! Thanks.
I hadn’t read this thread recently… It’s a great advancement.
It reminds me of my prompts in Antigravity, where when I ask for an action, it starts with a request to determine the appropriate tool for my request, then proceeds with those tools. Gladys is gradually becoming an agent! 
Everything is available in Gladys Assistant 4.84: Gladys Assistant 4.84 : Les intégrations externes sont là 🚀
I tested it and it works well!
However, I find the labels € economic · €€ medium · €€€ premium a bit confusing. A user might think each question will cost them money 
Hello,
I would like to connect my Gladys to my local server running on a gwen3.5 122B. Is there already a procedure described for this?
I could give you feedback on this configuration (admittedly a bit expensive, but local).
Is there a test protocol in place to compare the different models with each other?
Hello @l.dolle
I’m very interested in your feedback, especially since I’m currently getting quotes for a professional project involving an IA VLM + LLM server! I’m really intrigued to see Qwen3.5-122B running locally.
If it’s not too indiscreet, could you provide some details about the infrastructure? Type of machine, amount and type of RAM (unified like Mac/Strix Halo, or classic DDR5 + GPU?), quantization used, and the inference engine (Ollama, llama.cpp, vLLM…).
And most importantly, if you have figures from different angles: the time before the first token and the generation speed, with and without the thinking mode. That’s probably where it counts for home automation where you expect an almost immediate response.
Thanks in advance!
EDIT: For my part, I’m currently running small models on my pro laptop with LM Studio, which is very convenient for testing!
Hello @Terdious
The server is an MGKtec evo2 equipped with an AMD Strix Halo and 128GB of unified RAM.
I removed Windows (I didn’t see the point of having a local server if it was just to send all my data to Microsoft) and had to install Fedora Server (which uses Linux kernel 7.1) because Debian 13 only uses Linux kernel 6.12 and therefore does not recognize the Strix Halo.
I installed the entire AI environment with ODS (Osmantic Dream Server), which I invite you to check out.
Here is a screenshot of my system at rest.
I hope I’ve answered your questions.
Don’t hesitate if you need more information.
P.S.: I lack test protocols to evaluate my server. If you have any, I’m interested.
Hi @l.dolle, this is not possible at the moment, but I think we could add an API to external integrations so that it could become an external integration!
I wrote a request: Intégrations externes : pouvoir créer des intégrations de type "fournisseur IA"