Player not loading? Watch on YouTube
This tutorial builds a local AI browser-testing assistant with Qwen3:8B running through Ollama, LibreChat as the chat interface, and Playwright MCP controlling Chrome. The speaker presents the setup as running on your own machine without an OpenAI API key. The demonstrations access public websites, so local execution does not mean offline browsing.
The stated requirements are at least 16GB RAM and roughly 20-25GB of free disk space. An SSD and GPU are recommended; the GPU is optional. The speaker uses a Mac with an M2 Pro and 16GB RAM and says the setup also works on Windows and Linux. Prerequisites include Docker Desktop, Node.js 20 or higher, Git and Ollama.
The walkthrough covers downloading the model, configuring LibreChat, and starting its Docker Compose services. It explains why the container uses host.docker.internal to reach Ollama on the host. For Playwright MCP, the speaker uses the SSE endpoint on port 8931 after reporting failed attempts with the other transport.
The AI agent receives browser tools and instructions for sequential actions, fresh page snapshots and locator choices. Examples check the Playwright website's title and log into ParaBank with demo credentials. The speaker reports successful checks, but the demonstration also includes retries and long waits on the 16GB machine. The initial test produces mistakes before the agent instructions are added.
