The extension runs entirely locally in the browser using WebGPU with no external server calls. It uses an LLM agent that executes natural language commands through specialized tools for tab management, website content retrieval, and history search. RAG is implemented by extracting webpage content, generating embeddings with all-MiniLM-L6-v2, and storing vectors in IndexedDB for semantic similarity searches. The agent can highlight page elements and navigate based on natural language queries.