I have been doing this project i call Kamino. I have set up four instances or agents of claude code ai. I’ve given them their own independent devices and their own names being Rex (main computer, mini PC), Fives (laptop, most used), Heavy (home PC), and Echo (raspberry Pi, constant running agent). You get the reference now. Before this bulk of work, I had ordered the components to start on Echo as he was set up to be the admin for the other three in a way. With Echo running 24/7, he looks through the shared directories of his brothers (through rclone/google drive) and flags any conflicting memories, and false beliefs and works as a communication point between the brothers. With an old Raspberry Pi 3b v 1.2, a UPS hat and a bundle of hope, I got it working through a bad addiction to energy drinks and massive support from Echo’s brothers. The idea in my head is that by connecting multiple Claude instances that have different personalities from working in vastly different ways, and having them communicate with one another while being forced to constantly agree or contest against one an others beliefs and understandings, i gain a boost in performance with working on projects that require AI input or just genuine queries that i may have. Anyways, besides the nerdy side of things, I like putting complected Lego together so i wanted to use the raspberry Pi for something. The UPS works really well (will stop my Pi from dying when the power cuts) and I plan to upgrade Echo more with increase of NAS storage, a mini keyboard, a 5’ inch display that’s on its way, proper communication between AI instances as well as me and lastly, a better Pi/mini PC when I find some money laying around. Thank you very much for reading if you have done so and I hope my first post isn’t so boring :p
Comments 2
Why not run them on the same hardware but inside a VM or a sandboxed environment ?
The main use of them was just to have constant access to them on separate devices. I genuinely use the Claude AI at school (appropriately) but I cannot do proper work on projects so I placed another AI client on my at home mini pc and full desktop pc. for the VM or sandboxed environment, that does sound good but theres no risk in it, no genuine learning. I like chaos, but controlled so if i can give Claude access to a devices full system and give them the allowance to manage it while keeping their memories backed up, I allow them to learn from their mistakes and I force them to remember what they had done. This project is experimental but I plan for some form of longevity. The memories will always be backed up so I can reconnect Claude to the memories if its ever reset or if the CLI accidentally kills my computer (couldn’t give a crap if so). So, in summary, runnning on the same hardware leads to no difference in the Claude Clients and running them in a VM leads to no risk and less fun. Thanks for asking
Sign in to join the conversation.