DSH-Patrol

Having a large model re-plan every web or desktop interaction from scratch is slow and fragile: buttons may have moved, sessions may have expired, and compounding guesswork across many steps eventually drifts. DSH Patrol takes a different path. You describe the patrol workflow in natural language once, the Agent verifies and freezes it into a deterministic Runbook, and a lightweight Runner then replays that Runbook step by step without re-relying on the model to re-derive the path every time. Teach once, patrol repeatedly.

Capabilities span two axes. On the browser side, DSH Patrol auto-detects Chrome, Edge, or Chromium, launches an isolated persistent profile with a built-in extension, and keeps cookies and login sessions alive across Harness restarts. Credentials are stored only as reference IDs and never as plaintext. When a step requires a human token, a QR code, or a second confirmation, the run pauses and resumes the same execution once you are done. On the desktop side, Windows UI Automation is the primary primitive, with hotkeys, Windows OCR, and coordinate clicks as fallbacks. Built-in app knowledge bases cover WeChat, WPS, and Baidu Netdisk, and a Runbook can hand a web screenshot to a desktop app for notification or archiving, completing a cross-application pipeline.

It is built for teams and individuals who maintain internal ops dashboards, verify business-system web states on a daily cadence, need repeatable logged-in patrols with screenshot evidence, or want to chain a web-check into a desktop notify-or-archive step and have it replay deterministically every time.