Preface¶

DeepSeek Harness (dsh) splits the agent runtime into a plugin composition: the official repository positions itself as “everything is a plugin”, and the daily entry point is npx @deepseek-ai/dsh web or pnpm dsh web from the source code. This path works for those already familiar with Node.js and the command line; for more users who want to fit their workspace, plugins and sessions into a single window, the barriers are concentrated in three things: the local machine needs a runtime installed first, plugins are scattered across npm and GitHub and hard to find, and after installation, you have to check versions, permissions and compatibility on your own.

The community catalog lists deepseek-harness-studio under the “Notifications & Integrations” category. The repository is maintained by fufankeji, and the README states that the producer is Fankongjian (Normed Space). It does not add another notification channel to the existing dsh process, but uses Electron to combine the DeepSeek Harness Web workspace, local Host and desktop window, allowing plugin discovery, installation, Presets and workspaces to be completed on the desktop. This article is organized after cross-checking the community catalog page, GitHub repository README, Releases and the official deepseek-ai/deepseek-harness.

What is this¶

deepseek-harness-studio is an open-source desktop client for DeepSeek Harness. The one-sentence description from the community catalog is: No-code desktop client, one-click launch, supports Windows and macOS; built-in plugin discovery, popular plugin push, one-click installation and management, AI intelligent recommendation and visual enhancement.

The repository address on GitHub is fufankeji/deepseek-harness-studio, the main language is TypeScript, and the license is MIT. As of this review, GitHub shows 239 stars. It builds a desktop layer on top of the official deepseek-ai/deepseek-harness Harness core, Cordis plugin system and Web interface: this repository is responsible for the Electron entry point, local Host lifecycle, desktop interaction and supporting scripts, and does not replace the official Harness.

Two points need to be clarified first. First, the community plugin directory deepseek-harness-plugin.com is an independent site and has no official affiliation with DeepSeek / Fang Information Technology (Huankeji), and should not be treated as an official app store. Second, the “Normed Space Official” in the README only means that it is built and maintained by the Normed Space desktop development team, and does not represent the official DeepSeek Harness.

Core Features¶

The repository divides its capabilities into “Supported” and “Planned”. Below we only expand on the parts marked as available in the README; planned items will be listed separately in the notes to avoid treating the roadmap as ready-to-use features.

Desktop Window and Local Host¶

Studio uses Electron to host the DeepSeek Harness Web workspace, and the desktop main process is responsible for starting, waiting for and shutting down the local dsh web service when exiting. The window side provides system tray, single instance operation, external link handling and preload communication interfaces. The Web workspace still retains Harness capabilities such as sessions, workspaces, models, tools, Skills and plugins, just wrapped in a desktop shell.

The development preview packages downloaded from GitHub Releases come with the required runtime environment, and the README states that no additional installation of Node.js or pnpm is required to run the desktop client. When you need to modify the source code, you can still clone the full repository and launch it locally.

Plugin Discovery, Intelligent Recommendation and Plugin Hub¶

When you don’t know what to install, go to “Plugin Discovery” from the left sidebar. The app will read the public directory, display plugins by featured, recently updated, popular in the ecosystem and scenario categories, and also support searching by name, function or author. The scenario categories cover Agent & Workflow, Web UI, Browser & Search, Vision & Media, Memory & Context, Model & Service, Development Tools, Integration & Notifications.

When you only know your needs but don’t know the package name, you can hand the natural language to the current Agent. The app will send it as a /find-plugins request, and the Agent will only query the public dsh-plugin directory, returning candidate package names, versions, authors, update times and matching reasons. The recommended results are only directory metadata, and the actual installation still needs to go through the Plugin Hub for compatibility checks.

The Plugin Hub targets packages with the dsh-plugin tag on npm that comply with the DeepSeek Harness Bundle specification. Before installation, it will verify the confirmed version, product integrity, Bundle declaration and local machine compatibility; after confirmation, it will download and restart the Harness Host for runtime verification. The installed area can be viewed by system, public directory and local source, and you can enable, disable, update or uninstall via the menu. Uninstallation retains configuration and plugin data by default, and you need to confirm again to clean up the data.

Preset Plaza and Six Built-in Workflows¶

The README divides capabilities into three layers: Skill teaches Agents “how to do it”, Plugin / MCP provides executable tools, and Agent Preset bundles roles, Skills, tools and operating rules into a reusable working mode. The Preset Plaza is at the same level as Plugin Discovery and the Plugin Hub, and the process is: Discover → View capability composition and prerequisites → One-click installation → Select “Use for new session” in “Installed”. Installing or deleting user Presets does not require restarting the Host; system Presets are protected. The browser development mode is only for viewing the interface and will not modify the local Presets.

There are currently six built-in real-world workflows, totaling 6 Agent Presets, 8 Skills and 1 image generation Plugin. Feishu Digital Employee additionally connects Feishu MCP and time parsing MCP, and PPT Office uses the built-in animation runtime adapter:

The installation package delivers the streamlined Presets, Skills and tool adapters required for operation, and does not write case source code, input data or generated finished products into the user environment.

App Hub and Visual Enhancement¶

The App Hub is used to launch full AI applications with independent interfaces, exclusive data and their own operating processes, rather than identifying technical packages in the plugin list. The first built-in app is FF–LLM Wiki: it strings together document parsing, Wiki compilation, knowledge graph, vectorization and traceable question answering, managed by Desktop for local runtime environment, and opened in the system browser. Data is written to the app-specific directory under DSH Home, and does not mix into the current project workspace. The app does not carry the Normed Space team’s API Key; knowledge compilation, browsing and local retrieval can be run first, and you need to configure your own DEEPSEEK_API_KEY in the Harness credential center before using DeepSeek generative RAG.

The DeepSeek text model itself cannot directly understand images. After enabling “Visual Enhancement”, the built-in Bailian qwen3.8-max will first read PNG, JPEG, WebP, GIF images in conversation attachments or the workspace, then pass the traceable recognition results to the Agent, without replacing the current DeepSeek main model, permissions and session process. The first enablement requires verifying the Bailian API Key with a real image, and the credentials are only stored in the local protected credential file.

In addition, the input area provides Chinese permission levels (Read-only, Workspace Write, Full Access) and thinking modes (Turn off thinking, Deep Thinking, Maximum Thinking). General settings only determine the default permissions for subsequent new sessions; you must confirm the risks before enabling full access. The appearance can be switched in “Settings → Background” between “Official Original”, “Big Fat Fish Anthropomorphism”, “Cloud Cat”, or you can choose a local image; custom images are cropped locally to 1920×1080 WebP and color-matched, and the original image is not uploaded.

Installation and Activation¶

The installation command given in the community catalog is as follows, run in an already functional DeepSeek Harness terminal:

dsh plugin add github:fufankeji/deepseek-harness-studio

For reproducible installation, the catalog page recommends pinning the commit hash:

dsh plugin add github:fufankeji/deepseek-harness-studio#commit

Replace #commit with the actual commit hash. This command follows the dsh GitHub plugin installation path: the CLI will parse the repository and write it to the current profile. The repository itself is a complete desktop development environment, and the daily use path is mainly the installation package in GitHub Releases, do not mix it with the desktop preview package below.

Download Desktop Preview Package¶

The README states: Desktop installation packages are only released via GitHub Releases of this repository, and no third-party download stations are used. As of this review, the latest preview is DeepSeek Harness Studio 0.1.0-rc.9 (Pre-release), providing ZIP for macOS Apple Silicon and Setup.exe for Windows x64. Currently, all are unsigned development previews; the official version plans to provide signed macOS .dmg and Windows x64 .exe.

macOS arm64¶

1. Download and extract the preview ZIP from Releases.
2. Drag DeepSeek Harness.app into the “Applications” directory.
3. The current preview package has not been notarized by Apple, before opening for the first time, run in the terminal:

xattr -dr com.apple.quarantine "/Applications/DeepSeek Harness.app"
open "/Applications/DeepSeek Harness.app"

If the app is not placed in the “Applications” directory, replace the path with the actual location. This command only applies to preview packages downloaded from this repository’s Releases and verified with SHA-256. The release notes also state: Do not download the GitHub auto-generated Source code.zip / Source code.tar.gz, that is the source code, not the installation package. Currently, the macOS package only supports Apple Silicon (arm64).

Windows x64¶

Download DeepSeek-Harness-Desktop-Windows-x64-0.1.0-rc.9-Setup.exe and run the installer directly. The preview installer is not signed with Authenticode, and Windows may pop up SmartScreen; follow “More info → Run anyway”; do not turn off Defender for this. The 0.1.0-rc.9 notes mention: This version fixes the startup failure caused by the content directory not being created when FF–LLM Wiki starts for the first time.

Launch from Source Code (Development)¶

Downloading the source code, installing dependencies and launching the desktop development environment does not require pre-filling API keys. When you actually call the model, configure the selected service and credentials in the settings, and do not commit the keys to Git.

Environment requirements:
- Node.js ^22.19.0 || >=24.0.0
- pnpm 11.7.0

git clone https://github.com/fufankeji/deepseek-harness-studio.git
cd deepseek-harness-studio
pnpm install
pnpm run dev:desktop

Use pnpm run dev:desktop:rebuild when you need a forced full rebuild.

Typical Usage¶

The following items all come from the desktop operation paths specified in the repository README, and can be reproduced via the interface, not made-up cases.

1. Find and install plugins. Open the desktop client, go to “Plugin Discovery” from the left sidebar, browse by featured / recently updated / scenario, or search directly by name and author. Uninstalled plugins go through the secure installation process; installed ones are redirected to the Plugin Hub for management. You can also describe your needs in natural language on the discovery page (the acceptance example in the README is “Find a desktop pet plugin”), review the candidates and matching reasons, then go back to the Plugin Hub to confirm the installation.
2. Start a new session with a Preset. Enter the Preset Plaza, search by purpose or author, open the details to check the role, Skills, tools and external dependencies, and click install. Desktop will verify the source, size, summary and archive path, then write to the local user Preset directory. Go to “Installed” and click “Use for new session” to start the task with this working mode.
3. Open FF–LLM Wiki. App Hub → Select FF–LLM Wiki → “Open App” → Import materials, build Wiki and knowledge graph → Check sourced answers in intelligent Q&A. The left quick entry is closed by default, and can be turned on in “Show in sidebar”.
4. Add image reading to the current session. Open “Visual Enhancement” on the left side of the input box, follow the prompts to verify the Bailian API Key with a real image. Afterwards, you can hand screenshots, error interfaces, charts or workspace images to the Agent; the main model is still the current DeepSeek text model.
5. Change permissions and thinking depth. Select Read-only / Workspace Write / Full Access and Turn off thinking / Deep Thinking / Maximum Thinking in the input area. Full access will apply to the current session, check the risk prompt before enabling.

Applicable Scenarios and Notes¶

It is more suitable for these types of usage:
- Users who want to avoid touching Node.js and the command line, and directly use the window to manage DSH workspaces, sessions and plugins.
- Users who need to search and verify from the public dsh-plugin directory before installing, instead of manually piecing together GitHub addresses.
- Users who want to use ready-made Presets (Web products, HTML presentations, videos, graphic and text cards, Excel reports, Feishu tasks) to start new sessions.
- Users who still use the DeepSeek text model as the main model, but occasionally need to turn screenshots or workspace images into traceable visual observations.

Note these boundaries before use:
- Platform. The current preview packages are for macOS arm64 and Windows x64. The README does not list Linux desktop installation packages as released assets.
- Preview Status. 0.1.0-rc.9 is an unsigned Pre-release. macOS may be blocked by the quarantine mechanism, and Windows may encounter SmartScreen. Officially signed installation packages are not currently available as downloads.
- Planned capabilities that have not been released as usable features include: independent MCP / Skills / tool management, custom Agent and multi-Agent collaboration, task planning, background running and session recovery, project rules, Hooks and long-term memory, Git / Worktree and code review, browser and desktop automation, and mobile remote control and messaging channels. These items are marked as planned in the README, and this article does not treat them as usable functions.
- External dependencies. Video generation requires local FFmpeg and ffprobe; content factory image generation requires the logged-in Codex CLI with ImageGen permissions; Feishu Digital Employee requires Feishu app credentials; visual enhancement requires Bailian API Key; FF–LLM Wiki’s generative RAG requires your own DEEPSEEK_API_KEY.
- Permissions and Sources. The community catalog states: Plugins run with the permissions of the current dsh process, and may execute code during installation. Whether you use dsh plugin add or install the desktop preview package, you should check the source code repository and license before installation; the desktop package only gets it from this repository’s Releases and verifies the SHA-256. Do not run xattr to remove quarantine on unknown source installation packages.

Summary¶

deepseek-harness-studio wraps the DeepSeek Harness Web workspace and local Host into an Electron window, and supplements the desktop side with plugin discovery, Plugin Hub, Preset Plaza, App Hub and visual enhancement. It is maintained by fufankeji / Normed Space, licensed under MIT, and builds a desktop layer on top of the official Harness, but is not an official DeepSeek release.

Catalog page: https://deepseek-harness-plugin.com/zh-CN/plugins/deepseek-harness-studio/

GitHub: https://github.com/fufankeji/deepseek-harness-studio