computer-user
Run the following command in DeepSeek Harness:
dsh plugin install jing-hy/computer-user
Paste the following prompt into your AI chat to install this plugin:
Run dsh plugin install jing-hy/computer-user to install; the full source code is available at https://github.com/jing-hy/computer-user .
About this plugin
A pure text LLM cannot see a screen or drive a cursor out of the box. computer-user bridges that gap for DeepSeek Harness: any text-only model can screenshot the desktop, read it back via local OCR, and then click, type, drag, scroll, or press key chords on Windows — no multimodal model and no external vision API required.
The full loop is covered end-to-end: multi-monitor DPI-aware capture, UTF-16 CJK typing via Win32 SendInput, key-chord shortcuts, mouse drag, wheel scroll, cursor read-back, and timed waits. Paired with the local image_scan and image_ocr tools in picturereader, the model follows a locate-window, in-window OCR, click-once, screenshot-verify workflow. A settings card offers four safety modes (disabled, read-only, manual approval, automatic) plus an output guard that strips fake tool-call markup from the model stream.
Best for anyone running DSH on Windows who wants automated desktop tasks — form filling, UI inspection, or fully local GUI workflows — powered entirely by a text model and zero cloud calls.
Use Cases
- Drive desktop forms and app interactions with a pure text-only LLM
- Build a fully local GUI inspection and screenshot-analysis pipeline
- Let a model see and operate the Windows desktop without multimodal vision
Best For
- DSH users on Windows who need automated desktop operations
- Developers constrained to text-only models with no external vision API
- Builders pursuing fully local, zero-cloud-call workflows
Related Plugins
A method pack that makes AI coding agents plan against your real baseline, prove completion with fresh evidence, and reduce reworks and unsafe changes.
Turns the DeepSeek Harness session into a captain that builds a durable sub-agent team, splits goals into dependency-aware tasks, and coordinates work via direct messages and a live Web UI.
Gives coding agents design judgment, letting Claude Code, Cursor, and 70+ agents generate and iterate high-quality UI, presentations, and graphics right from the terminal.
Run the Pi ecosystem's plugins on DeepSeek Harness, unmodified, via a compatibility layer that implements Pi's public extension ABI on DSH's native services.