dsh Browser Control

dsh Browser Control connects DeepSeek's open-source, plugin-based DSH agent harness to the Chrome or Firefox tabs a user already has open. It lets the model read page content, operate controls, navigate and manage tabs while keeping the user's existing login state, session and cookies intact, with a side panel providing the conversation interface.

Browser operation stays entirely text-based: a page is converted into structured text with a numbered inventory of interactive elements, and the model addresses those elements by number rather than by looking at a screenshot; separate multimodal chat in the side panel can accept image attachments, but the browser-control channel itself never captures one. The project reports that, in a paired benchmark of sixty runs, its approach completed every run successfully while needing fewer tool round trips and less time than a comparable Playwright-based backend.

It installs with a shell or PowerShell script that also wires up the browser extension, and is built as a companion DSH bridge plugin and browser extension in one pnpm workspace, requiring a minimum DSH runtime version.