The 10 Plugins That Turn DeepSeek Harness from Bare Shell into a Workstation
On August 14, just a few days after DeepSeek Harness was released, I wrote a step-by-step installation guide.
Back then it had just come out. It could chat, it could work, but the interface was basically a bare shell.
Half a month later, the situation has completely changed.
The community plugin ecosystem has taken off, and many bloggers have done real-world cross-evaluations, ranking the plugins on the market. I went through the top 10 against that ranking one by one and compiled this list.
Below, I'll go through them one at a time. What each plugin does, who it suits — explained clearly in one go.
01 dsh-market, the ecosystem entry point
This plugin ranks first for a reason: it is the entry point to the entire ecosystem.
Install command:
dsh plugin --profile web add dshmarket
Once installed, restart dsh web, open Settings → Plugin Marketplace. From there you can search and one-click install community plugins and themes directly in settings, and also back up and restore configurations. Most of the 9 plugins covered next can be installed from here.

It's essentially giving the system an app store first. So no matter what you want to install later, step one is always to install this.
Repo: https://github.com/dsh-market/dsh-market
02 modlens, giving the model eyes
Universally recognized as the strongest vision extension.
DeepSeek's main models are text-only; if you paste a screenshot, it can't see it. modlens fills exactly that gap. Once installed, just paste an image into the input box, and it automatically outputs OCR text plus layout analysis.

Table screenshots, paper screenshots, error message screenshots — from now on no need to type them out manually, just paste them in.
Install command:
npx -y @deepseek-ai/dsh plugin --profile web add @liustack/[email protected]
Repo https://github.com/liustack/modlens, 3.7k stars, the hottest vision plugin in the entire ecosystem.
03 dsh-web-ui, interface all-in-one suite
Task kanban, Git graph, real-time token stats, even a desktop pet.
It upgrades that minimalist chat box directly into a full-featured console. You can see what step a task has reached, what the code repository looks like, how many tokens have been burned.
People who like to work while watching the interface, install this.

Install command:
dsh plugin --profile web add @linxin666/dsh-web-all@latest
Repo https://github.com/zhu1090093659/dsh-web, 6.4k stars, currently the highest-starred project in the DSH plugin ecosystem.
04 DSH-better-sidebar, side workbench
This one is for people who write code.
It recreates the VS Code experience, a full sidebar with built-in file tree, a real terminal, and sub-agent status monitoring. The sub-agents you dispatch inside Harness — you can see at a glance in the sidebar which step they're on now.

Pick one between this and 03, or install both, depending on personal taste.
Install command:
dsh plugin --profile web add dsh-better-sidebar@latest # First run will fail because pnpm 11 blocks the node-pty build script (dependencies are already written)
cd ~/.dsh/profiles/web && pnpm approve-builds --all # Allow the build script (auto re-runs install)
dsh plugin --profile web add dsh-better-sidebar@latest # Re-run and it succeeds
Repo https://github.com/omdsh-dev/DSH-better-sidebar, 3.1k stars.
05 dsh-at-file, quick referencing
Solves a high-frequency pain point.
Before, if you wanted the model to see a local file, you had to open the file, select all, copy, and paste a big chunk in. With this installed, just type @ in the input box, fuzzy-search local files and directories, hit enter to reference them into context.

Anyone who has used the @ function in Feishu or DingTalk gets it instantly.
Install command:
dsh plugin --profile web add https://github.com/omdsh-dev/dsh-at-file/archive/refs/tags/v0.6.8.tar.gz
Repo https://github.com/FSMargoo/dsh-at-file
06 deepseek-harness-desktop, native desktop client
A desktop client for the DeepSeek Harness official Web UI.
Download and install to use, no need to start dsh web yourself.
No Node.js environment configuration needed, download-and-go for both Windows and Mac. After installation it stays in the system tray; click to open whenever you need it.
For anyone who struggled with the environment setup originally, this skips that step for you directly.
Repo https://github.com/ChisaAlter/Deepseek-Harness-Desktop
07 dsh-web-search-pro, enhanced multi-source search
The default search capability is limited. This plugin aggregates DeepSeek, Bing, and Exa engines together, and also supports targeted crawling of platforms like Xiaohongshu, Zhihu, and Bilibili, with results that can be cached locally.
People doing research, gathering materials, or writing drafts — this saves a huge amount of time.
Repo https://github.com/anweat/dsh-web-search-pro
08 OpenViking, cross-session memory
By default, each session is isolated. Close the window, and everything you chatted about is gone.
OpenViking builds a self-evolving context database, turning fragmented conversations into long-term memory and knowledge RAG, so the model understands you better the more you chat. The project background you discussed with it last week — pick up the conversation this week, and it still remembers.
The engine itself is at https://github.com/volcengine/OpenViking. For the DSH-side integration plugin, the one the community uses most is https://github.com/Rxiain/dsh-openviking
09 dsh-cost-meter, cost dashboard
This plugin appeared at exactly the right time.
Everyone still remembers DeepSeek's price hike on August 17 — cache-hit pricing went from 0.025 to 0.3 during peak hours, a 12x jump. Running Harness now without watching costs is running blind.
dsh-cost-meter tracks session spend and daily budget in real time. The most practical feature: it automatically alerts you before DeepSeek switches between peak and off-peak pricing. Running tasks during discount hours saves quite a bit over a month.
Repo https://github.com/Han-1413141/dsh-cost-meter
10 dsh-session-library, knowledge base and deep reading
Last on the list, handling knowledge management.
Local document hybrid search plus long-text deep reading. It can automatically break down a very long PDF into claims and evidence chains — people reading papers and reports will love this.
Paired with 08's memory system, this turns Harness into a personal knowledge workstation combo.
A separate note on this one. The name dsh-session-library — I couldn't find a corresponding repo on GitHub; the author may have renamed it later. The closest in functionality is https://github.com/1-CellBio/dsh-okf, which similarly breaks research PDFs into a searchable knowledge base. You can also open dsh-market and search directly for session library.
Finally, how a newcomer should start
Finally, don't install all ten at once.
Three steps. Step one, install 01 marketplace and 02 vision first. Step two, install either 03 or 04, one of the interface plugins. By this point, the basic experience is already a completely different thing from the bare-shell version. Step three, then look at whether you use search or memory more, and fill in the rest as needed.
Save this list first. The DeepSeek Harness ecosystem changes week by week right now; today's top 10 might see half replaced next month. But the most core ones — marketplace, vision, interface — will likely stay stable the whole time.