One Model to Rule Them All: Qwen3.5 122B on a Homelab for CF Agents | CF Weekly #85
Originally aired live on March 6, 2026.
Four RTX 3090s and a homelab: this week we explore running Qwen3.5-122B-A10B locally with vLLM and putting it to work for Cloud Foundry agents.
Topics include GPTQ Int4 quantization, the moe_wna16 setting, OpenCode, and connecting the model to a BOSH-deployed OpenClaw instance. We explore how local models fit into developer and platform engineering workflows, alongside the practical trade-offs of homelab hardware and cloud subscriptions.
This podcast edition replaces the livestream countdown with our recorded introduction. The full closing conversation and original sign-off are preserved, followed by our podcast outro.
Cloud Foundry Weekly hosts: Nick Kuhn, Nicky Pike, Darin Zook, and Keith Lee.
