The hardware you need is becoming less as time goes by and the models get better. Just access any free online AI like copilot, tell it what hardware you have and what you want to do (coding for example) and it will tell you the best open source model to get and how to set it up. Took me about a half hour to do that.
A good person on here to follow for info on local AIs is @tydynrain who is making great progress running local AIs and agents.
Wishing you all the best!
I need to have a look. My new PC has a GPU that may do something.
You may find the following video useful to understand what various hardware can run when it comes to local AI.
Source: YouTube
I understand a bit better now. I bought my PC this year, but the high cost of memory limited what I got. I also need to work out what I would actually do with these tools. Generating images, video and music doesn't really interest me, but I might use them for coding.
How much VRAM does your system have, my friend? Using llama.cpp, or one of its forks like ik_llama.cpp, llama-cpp-turboquant, or buun-llama-cpp, it's feasible to run much larger MOE (Mixture of Experts) models than would normally be expected, since part of the experts are kept in RAM/CPU. Models like Qwen3.5 35B-A3B or Gemma-4 26B-A44B are very doable even with 6-8GB of VRAM. And as to coding, that's what I'm doing. I've been working on an autonomous liquidity-pool agent, to assist me in my asset-managing tasks. 😁🙏💚✨🤙
I have an 8GB RTX5060. I got it to help with things like video editing. I will look at this stuff when I find the time.
That's a very usable card. I have an RTX 4070 with 8GB of VRAM on my laptop. Perfect. Let me know if you've got any questions. 😁🙏💚✨🤙