Page Assist + Ollama: Gemma 4 browser setup and tests

Hugo Paredes walks through a Windows setup that connects Page Assist in Chrome to Gemma 4 4B through Ollama. He installs the extension from the Chrome Web Store, pins it, selects an installed model and sends a greeting to check the connection. In his setup, Page Assist finds the models while Ollama is running without additional configuration. He also notes that Ollama has macOS and Linux downloads.

The tutorial then tests the local LLM in two browser interfaces. In the side panel, Paredes enables access to the current page and requests a structured summary of Wikipedia's artificial intelligence article. He reports a 20-second response. In the full chat interface, he asks for an n8n contact-form webhook design with a JSON schema and validation rules, reporting 38 seconds. A final coding assistant prompt requests a security and performance audit of a short Python script, plus refactored code; he reports about 48 seconds.

The demonstrated system has an RTX 3070 with 8GB of VRAM and 64GB of system RAM. Paredes monitors GPU activity in Task Manager and says larger models that spill into system RAM run more slowly. These timings describe his examples, rather than general benchmarks. The workflow lets users run models locally, but the web-summary example uses an online page. The generated schema and Python changes are reviewed on screen; the transcript does not show execution tests that establish their correctness.