Palo | Devpost

Landing Page
Browser Version
Browser Version on GitHub

Inspiration

Today's browser agents are extremeley slow and token ineffiencient. We want to make browser automation more accessible by increasing the speed and efficiency of browser agents by reducing the amount of info that needs to be processed and sent to the model. We brainstormed for the project using Opennote.

What it does

Palo is a innovative browser that is faster and more efficient than all current market competitors (OpenAI and Anthropic). Palo extracts only the most relevant elements needed to complete a task. This browser allows for users to automate browser tasks such as searching, clicking, and form filling. We reduce token consumption by minimizing unnecessary context and execute tasks faster by simplifying how the agent interacts with the browser.

How we built it

We used SwiftUI to create an inuitive and Apple-native frontend interface, while using Rust for the backend. The backend uses websockets to efficiently and securely transfer page context to the model, mediating tool calls with the authenticated client user. The backend orchestrates cached web information, allowing it to seamlessly integrate cosine similarity search on vector embeddings and cache similar model outputs for immense speedups.