OpenClaw setup with Ollama and Qwen 3.5

This tutorial connects OpenClaw to a local LLM through Ollama, using Qwen 3.5 as the example. The presenter downloads a model, tests chat in Ollama, and configures OpenClaw to use it. Running inference on the computer avoids the cloud token charges discussed in the video; the demonstration does not establish that every connected service is free.

Memory determines which model fits. The presenter uses a Mac with 64 GB of physical memory and describes the 9 billion parameter model as requiring 6.6 GB, the 35 billion version as requiring 24 GB, and the 122 billion version as exceeding that computer's capacity at 81 GB. These are the figures given in the tutorial, rather than universal hardware requirements.

For a fresh OpenClaw installation, the walkthrough selects Ollama during setup, confirms its address and port, chooses local mode, and selects the downloaded model. For an existing installation, it uses the application command on Ollama's model page, then checks responses in the terminal and web interface.

The final section connects the AI agent to Zapier MCP. The presenter permits Gmail searches and draft creation, excludes sending and deletion, and tests retrieval of five emails. Zapier sponsors the video. This external service connection is separate from running inference locally.