OpenClaw + Ollama: local agent setup on Ubuntu

This tutorial walks through an OpenClaw and Ollama setup on Linux. Ollama runs the local LLM, while OpenClaw provides the agent interface for requests in plain English. The presenter checks Ollama version 0.23.2, downloads Llama 3.2 3B, and tests it with a short greeting before starting OpenClaw onboarding.

The setup uses Quick Start, selects Ollama as the provider, and chooses local-only mode. The presenter skips messaging channels, web search, skills configuration, and hooks, then opens the dashboard. OpenClaw's onboarding notice describes it as a beta hobby project and says shared or multi-user use requires lockdown.

The stated requirements are at least 8 GB RAM, about 10 GB of free disk space, and an internet connection for initial downloads. The demonstration machine runs Ubuntu 24.04 with 64 GB RAM and an Nvidia GPU with 8 GB memory. These specifications matter when assessing the response speed shown: the presenter switches to Qwen 3.6 and discusses the speed tradeoff between its 35-billion-parameter model and smaller models.

For the file task, the agent reports finding 15 images in Downloads and moving them into a desktop folder named 2026 without deleting files. The tutorial also covers logs, workspace Markdown files, and stopping and relaunching the gateway. The presenter calls this setup private and local-only. The demonstrated file task uses a local model, while initial downloads and any optional online services need network access.