Alpaca Browser is a Next.js based web application featuring an embedded Browserbase session managed by a Llama 3.2 90B instance hosted serverless on Together.AI. The front-end includes an intuitive sidebar chat, where users can prompt the LLM to carry out individual actions or orchestrate multi-step workflows to achieve complex objectives. Inspired by Anthropic’s recent computer-use demo, Alpaca Browser takes agentic web interaction to a new level, allowing for versatile, goal-oriented web tasks across a wide range of scenarios. We’re actively developing a customized dataset based on Mind2Web, capable of fine-tuning Llama’s capabilities, including its new vision features, for seamless interaction on common web interfaces through mouse and keyboard use—like we humans do. This fine-tuning will also empower Alpaca Browser to handle sequential tasks efficiently, enhancing usability and making it possible for users to accomplish more with minimal direct input. Due to time constraints, completing the fine-tuning entirely was not possible, but we’re looking forward to open-sourcing this version of Llama.