2026-04-29 20:27 UTC
@leetaur@social.librem.one Interesting. I'm using GPT-5.5 with OpenClaw. With 8GB VRAM, I haven't found a way yet to run OpenClaw with a local model served by Ollama. I can run a 7B model, but OpenClaw's context fills up the rest and leaks to RAM. Any idea how to set this up?
Replies (1)
-
@leetaur@social.librem.one 2026-04-30 02:15
@ebelo@mastodon.social To run OpenClaw and a good LLM locally, I would look at a Mac Studio, with a ton of RAM. Probably 128 GB RAM. Expensive! But that is my plan in the future. Right now I have OpenClaw running on my Linux box, but for the LLM I am pointing it at a cloud model LLM (Qwen 3.5) I first tried it on my M4 Mac Mini. I ran OpenClaw with Qwen3.5-9b completely locally, but it was painfully slow. I don't have my agents (castle residents!) talking with each other yet, but that sounds fun 🙂