junior steeple
Hi there!
I was wondering if anyone else is running OpenClaw with fully locally LLM models (no external ai services like Claude).
I’m currently setting up a small AI cluster that runs OpenClaw on a Mac Mini M2 (this also runs liteLLM as an OpenAI compatible “proxy” to the different models running on my sparks) and 2 Nvidia DGX Sparks connected with a 400Gbs QSFP-DD link, RDMA, vLLM, etcetera…
I have 4 Sparks in total, but waiting for more QSFP-DD cables to get everything connected into a 3 spark cluster with 384Gb unified memory for huge models (3x 128Gb) and an additional spark for smaller models and other tasks
I’m currently running Qwen 3.5 122b using 2 sparks with a large context window successfully and it’s awesome!
Anyone else having experience with an equivalent hardware setup to share some good tips?
supple igloo