Biotic Browser is an experimental AI system designed to rethink how people interact with the web through intelligent agents. Rather than treating the browser as a passive window into information, Biotic Browser turns it into an active collaborator, capable of understanding goals, navigating complex interfaces, and maintaining context across extended interactions.

The project explores what it means to work alongside an AI rather than issuing one-off commands. Inspired by the experience of being a passenger in an autonomous vehicle, users observe the system reasoning, acting, and adapting in real time. The interface makes the agent’s decisions legible, allowing users to intervene, redirect, or refine intent as tasks unfold.

At its core, Biotic Browser leverages long-context language models to maintain continuity across complex, multi-step tasks such as online research, form completion, or multi-site workflows. Early iterations explored multimodal approaches using vision-based page understanding, while later versions transitioned to DOM-level parsing to improve reliability, performance, and privacy. This shift enabled the system to operate without transmitting visual data, while preserving awareness of page structure and interactive elements.

A key contribution of the project is its emphasis on temporal continuity. Unlike conventional chat-based assistants, Biotic Browser maintains an evolving internal state, tracking prior actions, goals, and outcomes over extended sessions. This enables behaviors closer to genuine task execution rather than short-lived question answering.