alternative would be to get a “cheap” m2/m3 MacBook Air / Mac mini - those work pretty well too because they share the RAM between GPU and CPU and there is no separation
but I see a 12GB RTX graphics card (you need cuda support I think - otherwise the large gpu won’t help) should cost you <300€
with models ~20GB you get pretty decent results - but a bit above 8GB it starts to become interesting lately =)
(aaaaand … this now is dangerous half-knowledge …but if you have multiple graphics cards - even when they cannot officially be connected to one large gpu - Ollama should automatically split the layers of the llm to use all GPUs - so 2x12GB should be as powerful as 1x24GB … but ofc requires more space in your pc … additional slots … power …)
Great job @saykor !
And to @Dimitar ( ) for his persistent, enthusiastic and “just get it done” attitude !!
Seeing all this ideas pop up, after all this years, is fantastic.
Local model connection would obviously be very welcome, the privacy loop needs to be closed.
May I ask whether the demo runs storing data to a test local network? Or else?
Sync on start and exit - when the app starts, it downloads the database from the Autonomi network. When you close it, it uploads the updated version. This is how it’s currently done because ant-cli (the tool used to interact with the network) still lacks full capabilities. For example, it doesn’t yet support Scratchpad, which would let us make real-time changes without uploading the whole database. Once that’s available, we’ll switch to a more efficient method. For now, we just upload and download a small sqlite file (a few KB) - simple but effective.
They don’t have a native .NET API yet, so I have to use ant-cli. But as you can see, we’ve got notifications where the two apps talk to each other and stay in sync like gossiping coworkers.
In the next stage, I’ll be adding an option to switch to a local model - a mini AI that runs directly on the device without needing internet access. That way, even if the Wi-Fi disappears, Queeni still knows what you want… unlike some people.
There’s no documentation on whether it works on Windows or how to run it on a test network. It looks like you need to use both CLI tools - ant and mutAnt - at the same time.
I’m planning to add cross-plateform support right after the release of Mutant v0.6.0, which will hopefully happen before the end of the week
I can ping you when I have a working alpha if you want @saykor ?
👑 Queeni — Your AI Assistant, Now Smarter, Private & Decentralized!
Queeni has evolved. She doesn’t just search, remind, and organize anymore — now she thinks independently, works offline, and can tap into decentralized AI networks when needed.
🧠 New Capabilities:
📦 LocalAI & Ollama – Run AI models directly on your device. No cloud. No data leaks.
🌐 Decentralized AI (e.g. Corcel) – Tap into global intelligence, not controlled by any single company.
🚀 What This Means for You:
🔒 Privacy by default: Your data stays with you.
🧭 Freedom of choice: Switch between local and decentralized AI anytime.
📴 Offline mode: Queeni works even with no internet connection.
🛠️ No dependencies: Free from OpenAI, Google, or any centralized provider.
⚙️ Expandable: Add or train your own models. Connect to new AI networks.
🔮 Why It Matters:
Queeni isn’t just a tool — she’s a step toward AI sovereignty. In a world dominated by centralized systems, Queeni gives you control, privacy, and independence. She’s not just smart. She’s yours.
So I presume this will enable payment for AI use with token?
I’ve been using the Zed editor - which a couple of weeks ago added MCP server systems for full AI integration, which is really cool. I know some other editors have had this for a while (windsurf and visual studio for example), but Zed is written in RUST and is very fast and low resource. Anyway, my question is - will Queeni AI have an API that can be added for using ZED something like ZED? Or will it be an editor all on it’s own with MCP server capability to write and edit files/folders?
Cheers
Edit: Zed is open source, so perhaps could be forked into Queeni AI maybe?
Think of Queeni as a wrapper (and of all Autonomi programs in general, because your data is on the network and you can always use another app to access it).
In this sense, there is no technical obstacle to using any local/external AI, as for external ones you have to pay them directly, Queeni only takes care of the data records in Autonomi.
One advantage of combining network records with an AI assistant is that you can switch between a local and an external model.
The local one works with your already saved data, for example, you saved a task to take your grandmother to an ophthalmologist and after 2 months you need to take her again, but you forgot what his name is - you can use the local model to search for you in the tasks and find it, and the external model to find his phone number on the internet to make an appointment.
You changed your Android phone during these 2 months, no problem - you install Queeni on the new one and it has access to your data.
Right now, I have no plans to sell AI usage or monetize it - unless Queeni somehow ends up with millions of users overnight (which would be both amazing and mildly terrifying ).
As for integration with Zed or other editors - I totally get the interest, but Queeni isn’t heading in that direction. She’s meant to be a personal assistant, a smart interface that can connect to different AI models and help with tasks like reminders, suggestions, searching, and so on.
The cool part is that the same principle behind Queeni - AI performing meaningful actions in software - can be applied anywhere. But for now, she’s her own queen , not looking to move into someone else’s castle.
If there is an API, then it should in theory be relatively easy to integrate into the many editors (including ZED).
I assume from the OP that it’s going to run on the network … so if no token payment for API calls, then what is the incentive for people to run AI nodes to process API calls? Am I missing something here? Perhaps I am misunderstanding your model completely.
You’re missing the point that there are no computational processes in Autonomi for now.
Everything happens on your computer. Queeni is a program that you install on your computer and it interacts with local AI or a cloud-based one and saves the results in Autonomi so that you have eternal access to them.
My new video card (GeForce RTX 3060 Ti 8GB) is arriving next week. Right now I’m still rocking an AMD Radeon HD 5870 - but once the upgrade lands, Queeni will go wild testing local models!
gemma3:12b with 8.1GB is a little bit too large for it - but with cpu-offloading the last layers it should still run pretty fast! (faster than the online hosted solutions like chatgpt when using their web platform I think)
yeah - it really starts to slow down significantly the larger the cpu % of the llm is no matter what you do - but 8GB is a decent size and should allow for reasonable results at good speed
(at least I hope the windows version of Ollama supports cpu offloading while mainly using gpu - but then again you can always just activate wsl and run Ollama there while having everything else running in windows … it’s just an api …)
to 
) for his persistent, enthusiastic and “just get it done” attitude !!
token?
).
, not looking to move into someone else’s castle.
