跪拜 Guibai
← All articles
Artificial Intelligence · Open Source · DeepSeek

The 10 Plugins That Turn DeepSeek Harness from Bare Shell into a Workstation

By 狂师 ·
Read original on juejin.cn ↗ Google Translate ↗ Alt translation

DeepSeek Harness is gaining traction as an open-source AI coding agent, but out of the box it is text-only and session-isolated. This plugin stack closes the gaps that would otherwise make it uncompetitive against integrated commercial alternatives — vision, persistent memory, cost controls, and a proper IDE-like UI — and the marketplace itself signals that the ecosystem has reached self-sustaining velocity.

Summary

DeepSeek Harness shipped as a bare chat interface, but a fast-moving plugin ecosystem has filled the gaps in under a month. The essential first install is dsh-market, an in-app marketplace that lets you search and one-click install everything else. modlens gives text-only DeepSeek models OCR and layout analysis on pasted images, while dsh-web-ui and DSH-better-sidebar compete to turn the chat window into a full console with task boards, Git graphs, file trees, and live terminals. dsh-at-file brings fuzzy @-mention file referencing into the input box, and dsh-web-search-pro aggregates multiple search engines with platform-specific crawling for Xiaohongshu, Zhihu, and Bilibili.

On the infrastructure side, OpenViking builds a self-evolving context database so conversations persist across sessions as long-term memory, and dsh-cost-meter tracks per-session spend with automatic alerts before DeepSeek's peak/off-peak pricing switches — a direct response to the August 17 price hike that pushed cache-hit costs up 12x during peak hours. A desktop client eliminates the Node.js setup entirely for Windows and Mac. The recommended onboarding path is marketplace plus vision first, then one interface plugin, then search or memory as needed; the ecosystem is still turning over weekly, but marketplace, vision, and interface are expected to stay stable.

Takeaways
dsh-market is the mandatory first install: it adds an in-app plugin marketplace to DeepSeek Harness for one-click installs, theme switching, and config backup/restore.
modlens (3.7k stars) adds OCR and layout analysis to the input box so text-only DeepSeek models can process pasted screenshots, tables, and error messages.
dsh-web-ui (6.4k stars, the highest-starred plugin) upgrades the chat UI into a full console with task kanban, Git graph, real-time token stats, and a desktop pet.
DSH-better-sidebar (3.1k stars) replicates a VS Code-style sidebar with file tree, real terminal, and live sub-agent status monitoring.
dsh-at-file lets you @-mention local files and directories in the input box for fuzzy-search referencing instead of manual copy-paste.
deepseek-harness-desktop is a native desktop client for Windows and Mac that removes the need to configure Node.js or run dsh web manually.
dsh-web-search-pro aggregates DeepSeek, Bing, and Exa search engines and adds targeted crawling for Xiaohongshu, Zhihu, and Bilibili with local result caching.
OpenViking builds a self-evolving context database that persists fragmented conversations into long-term memory and knowledge RAG across sessions.
dsh-cost-meter tracks per-session spend and daily budget in real time and alerts users before DeepSeek switches between peak and off-peak pricing tiers.
dsh-session-library (or the functionally similar dsh-okf) breaks long PDFs into claims and evidence chains for hybrid local search and deep reading.
The recommended onboarding sequence is: marketplace + vision first, then one interface plugin (03 or 04), then search or memory plugins based on individual usage patterns.
Conclusions

The ecosystem reached critical mass in under a month — a marketplace, competing UI overhauls, and cost-control tooling all emerged, which is unusually fast for an open-source agent platform.

dsh-cost-meter is a direct market response to DeepSeek's August 17 pricing change (a 12x peak-hour increase on cache hits), showing that plugin developers are tracking the commercial API's economics in real time.

Two interface plugins (dsh-web-ui at 6.4k stars and DSH-better-sidebar at 3.1k stars) compete for the same slot, suggesting the community has not yet converged on a single UI standard.

The desktop client eliminates the Node.js environment setup barrier, which was a known friction point in early adoption — packaging an agent tool as a tray-resident desktop app lowers the entry floor considerably.

The list itself is compiled from cross-evaluations by multiple bloggers who ranked plugins, indicating a secondary information layer (plugin rankings and reviews) is already forming around the ecosystem.

Concepts & terms
DeepSeek Harness (DSH)
An open-source AI coding agent platform built around DeepSeek's models. It provides a chat-based interface for dispatching autonomous coding tasks, extended through a community plugin system.
dsh-market
The in-app plugin marketplace for DeepSeek Harness. Once installed, it allows users to search, one-click install, and manage community plugins and themes directly from the settings panel.
modlens
A vision extension plugin that gives text-only DeepSeek models the ability to process images pasted into the input box by automatically performing OCR text extraction and document layout analysis.
OpenViking
A self-evolving context database engine that converts fragmented chat conversations into long-term memory and retrievable knowledge (RAG), enabling persistent cross-session context in AI agents.
Peak/off-peak pricing (DeepSeek API)
DeepSeek's API pricing model where cache-hit token costs vary by time of day. On August 17, 2024, peak-hour cache-hit pricing rose from ¥0.025 to ¥0.30 per token, a 12x increase, making cost-tracking plugins economically necessary for heavy Harness users.
Source: juejin.cn ↗ Google Translate ↗ Backup ↗