dsh-browser-client

You ask the LLM, What does this page show right now?, and it can only guess from a screenshot you manually paste. dsh-browser-client drops a DSH chat panel into the bottom-right corner of your browser and gives the LLM a set of page_* tools to capture structured page snapshots, run JS, click, type, scroll, take screenshots, and inspect network requests-all against your real, already-logged-in browser, so cookies, SSO, and intranet permissions just work out of the box.

It goes beyond describing what it sees. When you describe a repeatable web task in plain language, the LLM enters an exploration mode: it tries actions, records successes and failures, verifies the goal, then distils only the validated path into a clean, re-runnable workflow. Failed attempts are discarded automatically. Workflows can be scheduled, registered as DSH skills, and shared via parameterised JSON templates covering six common scenarios-batch forms, data collection, content transfer, bulk operations, multi-account tasks, and inspections-so non-developers never face a blank canvas.

Built for business and ops teams who need LLM-assisted interaction with intranet systems, authenticated pages, or repetitive web workflows. If you can use a browser, you can delegate multi-step page operations to the LLM through a single conversation.