dsh-pilot: Driving Local Browsers in DeepSeek Harness Chat

dsh-pilot is a plugin for the DeepSeek Harness Web GUI, designed to drive native Edge or Chrome browsers via CDP to perform browser operation tasks. It supports opening pages, reading structured text, clicking and inputting based on snapshot references, filling forms, and uploading files, all without requiring an API key or runtime dependencies. Users can observe real-time screenshots, URLs, status, and operation logs through the cockpit panel. Each agent session uses an isolated browser instance, ensuring that parallel sessions do not interfere with each other. Installation requires Node >= 22, and the plugin can be added via the `dsh plugin` command from GitHub or a local directory, followed by restarting the `dsh web` process. Key limitations include support for only headless mode, single-session single-tab functionality, and restricted interaction capabilities.

Read More