Here’s my OpenClaw setup, and the best part is it doesn’t cost me anything extra because I’m running an open‑source LLM locally on my own machine. In this video I walk you through getting Ollama installed, grabbing a model (I’m using Qwen 3.5), and then wiring that straight into OpenClaw so you’re not paying per token like you would with Claude or ChatGPT. If you want OpenClaw running autonomously for as long as you like without an ongoing bill, this is the cleanest path. I also break down the practical side of local models: parameter sizes, memory requirements, and what you can realistically run on your hardware. On my Mac with 64GB of memory, the 9B Qwen 3.5 baseline is easy, and I can even step up to the 35B version if I want better outputs. Then I show two setup flows: a fresh OpenClaw install where you choose Ollama during the wizard, and an “already installed” path where you swap your config to point at the local model. Finally, I show how I connect apps and services to OpenClaw using MCP (with Zapier as the middleman) so you can reduce the cost and complexity of big context queries and API-style workflows—while still keeping granular control over what tools OpenClaw can access.