omdsh-dev/dsh-browser

What it solves

This project provides a bridge between the DeepSeek Harness (dsh) agent framework and a user's active web browser. It allows an AI agent to interact with the web as a human would—reading page content, clicking buttons, filling forms, and managing tabs—while maintaining the user's existing login sessions, cookies, and authentication states, avoiding the need for headless browser automation tools like Playwright.

How it works

The system consists of a bridge plugin for the dsh runtime and a browser extension (Chrome/Firefox). The agent interacts with the browser via a text-only interface: the extension converts the web page into a structured text snapshot containing a numbered inventory of interactive elements. The model then identifies and operates these elements by their assigned numbers rather than using complex selectors or screenshots.

Who it’s for

Developers and users of the DeepSeek Harness who want to give their AI agents the ability to operate a real browser in real-time, specifically those who need to maintain active user sessions and avoid the system overhead of headless environments.

Highlights

• Session Preservation: Operates within the user's actual browser tabs, retaining all logins and cookies.
• Text-First Interface: Uses numbered inventories of page elements for navigation and interaction, reducing the need for visual processing.
• Privacy-Centric: Automatically masks passwords and payment card information so sensitive data never leaves the page.
• Integrated UI: Includes a side panel for conversation and a quoting feature that allows users to highlight text on a page to send it directly to the agent.
• High Performance: Benchmarks show lower latency and fewer tool calls compared to traditional Playwright-based automation.