WorkDSH

Give AI the work. See the process. Keep the result.

An open-source desktop workspace built on DeepSeek Harness, bringing projects, documents, experts, skills, connectors, and tasks together.

Download Desktop · Explore the workflow · User guide · 简体中文

_{Captured from a local WorkDSH session. Project names and account figures are demonstration data, not bundled with the installer.}

Why WorkDSH

An AI task needs more than a prompt: it needs source material, working rules, the right capabilities, and a result people can inspect and keep editing. WorkDSH organizes these in a desktop workspace while retaining DeepSeek Harness's native models, tools, sessions, / commands, @ references, and attachments.

A common problem	How WorkDSH approaches it
Re-explaining the context in every conversation	Projects bring instructions, plans, tasks, material, and capability configuration together.
Files scattered across chats and directories	The library manages local files, search, and previews; tasks can reference material and its revision.
Useful working methods are hard to reuse	Skills hold callable instructions and resources; experts publish role and capability configurations as fixed revisions.
A chat answer still needs a separate delivery tool	The deliverable workspace previews or edits supported working copies of documents, spreadsheets, presentations, HTML, and PDF.

From material to deliverable

File in library ──reference──> Project task ──choose──> Expert / Skill / Connector
                                            │
                                            └──> Inspect the process and result; keep editing in supported editors

This path is WorkDSH's product direction. End-to-end validation of project document references, expert execution, and different Office formats is ongoing during Alpha. Preview, editing, and export fidelity differ by format; the current capabilities and limits describe them in more detail.

See an HTML deliverable from a real local conversation

_{A local task example showing deliverable cards and right-side preview; it does not imply lossless editing for every file.}

A path into a large Skill ecosystem and DSH plugins

WorkDSH uses two complementary kinds of extension; a skill is different from a plugin:

	Skill: reuse a working method	DSH plugin: extend the system
Purpose	Give an agent instructions, scripts, references, and resources	Extend the Harness Host, client, tools, or services
How to add it	Import a file or ZIP package containing SKILL.md, inspect it, then confirm installation	Use the official plugin loading and composition mechanism for the pinned DSH version
In WorkDSH	A local skill catalog supports search, enable/disable, and management; WorkBuddy-style and community Skills can be tried for migration	WorkDSH projects, library, experts, skills, and connectors are feature packages; third-party DSH plugins can be adapted to the pinned version

“Large compatible ecosystem” describes an open import format, not a huge preinstalled catalog. Many WorkBuddy-style skills use SKILL.md as their entry point. WorkDSH supports that structure and bundles a skill-creation guide inspired by WorkBuddy's full authoring workflow. Scripts, external services, permissions, and proprietary formats still need individual validation. Third-party DSH plugins must likewise be checked against the current upstream version; a community directory listing is not a compatibility test. Skill management · Plugin development · Ecosystem manifesto

_{The installable entries shown here come from the demonstration machine's local skill directory. They are neither bundled with the installer nor an officially hosted online marketplace.}

Download Desktop

The current public desktop installer release is 2.0.5-alpha.20. These links point directly to files in its GitHub Release:

Platform	Download
Windows x64	WorkDSH Setup.exe
macOS Apple Silicon	WorkDSH arm64.dmg
macOS Intel	WorkDSH x64.dmg

Ordinary users do not need to install DSH, Node.js, or Python separately; the desktop package contains pinned runtimes. The macOS DMGs are currently unsigned, and updates are downloaded manually from Releases. Start with the user guide and FAQ.

Technical foundation and current limits

WorkDSH uses a pinned, unmodified official DeepSeek Harness submodule. Electron handles the window, startup, and packaging. WorkDSH features are composed as plugins in one Profile; Desktop does not install a second DSH runtime. See upstream.json for the pinned version. Whether models or third-party services use the network depends on user configuration.

This is an Alpha preview. Real cross-platform tasks, long-running expert team execution, fidelity for arbitrary Office files, and the community plugin marketplace still need validation or implementation. Published desktop packages may differ from the development branch; consult each Release for its scope and assets. Report issues on GitHub.

Development and documentation

Source ownership: WorkDSH feature packages and Web · Desktop carrier · Architecture · All documentation. Running from source requires Node.js 22.19+ or 24+, Corepack, and Yarn 4.18.0:

git submodule update --init --recursive
corepack yarn install --immutable
corepack yarn dev

Run checks with corepack yarn check. Contributing

Community and acknowledgements

Thanks to DeepSeek Harness, Cordis, and the open-source community. Thanks to Alibaba Cloud Wuying Cloud Computer, UCloud AstraFlow, and 88API for supporting the project.

GitHub Issues · Discord · Contact maintainers

WorkDSH uses the MIT License. It is an independent community project and is not affiliated with, partnered with, authorized by, or endorsed by DeepSeek or WorkBuddy. Those names appear only to describe technical origins, compatibility, and design references. Upstream contributors shown on GitHub are inherited from synchronized commit history; this does not imply that they maintain this repository.

Star history