dsh-fish-memory: A Fading Long-Term Memory for DeepSeek Harness Sessions
`dsh-fish-memory` is a DSH plugin designed to address the limited context window problem. It simulates a “fish memory” mechanism, automatically filtering content by reducing data “freshness,” storing important information in local JSON files, and preventing irrelevant clutter from filling up the context. The plugin supports toggling session memory, managing factual keys, importing old archives, and visual analysis, making it well suited for AI assistants or companions that require long-term maintenance of contextual coherence.
Read Moredsh-t3-model-picker: A Model Picker Ported from T3 Code
`dsh-t3-model-picker` is a DeepSeek Harness Composer plugin that registers at a lower priority to override the default model picker. It ports T3 Code's `ProviderModelPicker` and provides features such as a provider rail on the left, search, favorites (stored in the browser), jump shortcuts (Cmd+1-9), and inference effort level display. Users can open the card to search and save favorites using the shortcut `Cmd+Shift+M`, but the shortcut only works while the card is open and does not apply to subagent sessions. The installation command is `dsh plugin --profile web add dsh-t3-model-picker`.
Read MoreNexus for DSH: A Native Coding Task Assistant for Desktop and Messaging Channels
Nexus is a plugin built on DeepSeek Harness that integrates local model execution with remote collaboration workflows. Through task cards and a process sidebar, it connects local Codex or Claude Code to enable closed-loop management of requirement handoff, progress tracking, result review, and task continuation. The plugin supports desktop and multi-channel collaboration via WeChat, Feishu, and other channels, and provides native components for memory, calendar, email, and to-do items. After installing and configuring coding tools and mobile channels in DSH, users can schedule and accept coding tasks through a unified interface, effectively resolving the disconnect between local execution and remote collaboration.
Read Moredsh-intelligent-ui: Model-Generated Declarative DSL Interface
dsh-intelligent-ui is a plugin for DeepSeek Harness (DSH) designed to enable models to directly generate interactive declarative DSL interfaces. The plugin supports secure code execution in a DOM-free Worker sandbox and uses the host's native DOM for rendering, providing in-place rewriting, version history, and persistence across restarts. Installation requires DSH >= 0.2.1-rc.1. During use, the model mounts the interface via the `dsh-artifact` tag, making it suitable for generating interactive tools such as calculators and configuration panels.
Read Moredsh-stepwise-distill: Solidifying DSH Session History
`dsh-stepwise-distill` is a DSH plugin for addressing history bloat in long-conversation scenarios. It stepwise consolidates session history, rewriting tool results as the facts actually retained by the model, preventing repeated resending of completed content, and improving context density. The installation command is `dsh plugin --profile <name> add dsh-stepwise-distill`. Summary features (`stepSummary` and `turnSummary`) are enabled by default, and their status can be checked via the `/distill` endpoint. Note that this plugin roughly doubles token and time costs and requires a Node.js >= 20 environment.
Read Moredsh-kubejs: Customize Installed DSH Plugins with a Script Package
dsh-kubejs is a plugin for DeepSeek Harness (DSH) designed to mitigate the breakage and compatibility risks caused by directly modifying plugin source code through standalone script packages. It fully isolates modification logic from plugin code, supporting declarative configuration for event hooks, UI component registration, configuration overrides, and global fetch interception. Scripts are organized by target plugin, include dependency management and a configuration ledger, and provide hot reloading and a visual panel. Developers use the five primitives provided by `activate(api)` to implement customizations without touching the source code, ensuring the safety and maintainability of modifications.
Read MoreDSH GitHub Sync Plugin
The `dsh-github-sync` plugin securely syncs DSH plugin code to GitHub as a “clean copy.” It uses a two-stage safety confirmation mechanism, stores tokens locally only, and automatically excludes runtime files to avoid conflicts. Users must configure token permissions (such as fine-grained Contents and Administration) and user information. The plugin supports incremental synchronization, standard Git workflows, and natural-language-driven operation. It does not upload automatically; it is triggered only on demand and enforces a strict limit of 100 MiB for any single file.
Read Moreyueyexiayu/dsh-oauth: DSH Official OAuth Login Plugin
The DSH OAuth plugin provides official sign-in capabilities for Grok and ChatGPT Codex in the DeepSeek Harness desktop client. By integrating the official OAuth flow, it enables device-code interaction and state management, with tokens securely stored in the host-side configuration file. It supports account switching, token refresh, and session routing management, and includes robust error handling. Installation requires placing the files in the specified directory and registering them in the configuration file; changes take effect only after the application is fully restarted. The plugin is available only to users with the corresponding membership permissions.
Read Moredsh-audio-converter: A Local Audio Decryption and Conversion Tool
The “dsh-audio-converter” plugin for DeepSeek Harness (DSH) follows the “everything is a plugin” philosophy, designed for local audio processing without cloud services. It supports decrypting encrypted music formats such as NCM, KWM, KGM, and QMC, and enables conversion between multiple formats as well as audio quality enhancement. The plugin can be packaged as a single-file .exe, supporting drag-and-drop and batch processing. Installation is simple and depends on ffmpeg. Note that some formats (such as the new QQ Music version and KGG) have compatibility limitations, and use should be limited to personally purchased music.
Read MoreDSH Rail Music: Make Dialogue Rails Move to the Beat
Rail Music is a DSH Web plugin that captures system audio to drive the height of rails, giving navigation rails a rhythmic visual effect. It supports Windows, Linux, and macOS, and provides multiple visual styles and shortcut-key controls. Before using it, Python dependencies and specific audio driver configuration are required, and the effects of browser settings and exclusive output mode should be considered.
Read MoreDeepSeek Harness Web Plugin: Enter for Newline, Cmd/Ctrl+Enter to Send
`dsh-composer-modenter` is a Web plugin for DeepSeek Harness designed to address the inconvenience of multi-line input caused by the default Enter key triggering a send action in Web mode. Using browser-side logic, the plugin changes the Enter key behavior in the input box from “Send” to “New Line,” while preserving send with Cmd/Ctrl+Enter, slash menu selection, and input method compatibility. The installation command is `dsh plugin --profile web add github:yonchicy/dsh-composer-modenter`. The changes take effect after restarting the GUI, making it suitable for developers accustomed to code-editor-style interactions.
Read Moredsh-cache-warmer: DeepSeek Harness Plugin
`dsh-cache-warmer` is a core plugin for DeepSeek Harness (DSH) designed to optimize inference cost and latency through prompt caching. It provides cache time estimation, token usage monitoring, and cost-aware warm-up. After installation, users can configure model policies in the “Plugins → Cache warmer” menu, including setting cache lifetime and enabling warm-up. The plugin supports most text chat models, is disabled by default and must be manually enabled, and has been compatibility-reviewed for specific adapter versions. Developers should note the special configuration required for Codex/OAuth routing and the limitations of client-side cancellation.
Read MoreIntroduction to the DSH Prompt Optimizer Plugin
Users of DeepSeek Harness (DSH) often experience Agent execution deviations due to ambiguous input. The DSH Prompt Optimizer plugin aims to address this by transforming casual input into structured, executable prompts. Positioned between the user input and the Agent, it features a “user review mechanism”: optimized results are shown only in a preview popup and are written into the input box only after user confirmation. Installation requires DSH version 0.1.7-rc.2 and a pnpm environment. Its core features include reading drafts, intelligently analyzing the task domain, structured completion of constraints, and preview comparison. The plugin supports custom optimization models, thinking intensity (light/balanced/deep), and optimization language. It includes built-in strategies for domains such as Minecraft, UI design, and game development, and
Read Moredsh-vision-tool: Adding Image Understanding Capabilities to a Text-Only Main Model
dsh-vision-tool is a plugin designed to address the issue that DeepSeek Harness (DSH) cannot handle images with text-only models. By relaxing admission controls, registering tools, and injecting guidance, it delegates image understanding to a user-configured multimodal model. The plugin includes an intelligent backoff mechanism, so models that natively support images are not affected. Installation requires running the command in the DSH environment and being aware of the security risks related to pnpm build authorization. The configuration includes the vision model provider and its parameters; during use, users paste an image and ask a question, while the main model automatically invokes the tool and obtains the vision model's analysis results.
Read Moredsh-peak-scheduler: Automatic Peak/Off-Peak Pricing Scheduling for DeepSeek
`dsh-peak-scheduler` is a DeepSeek Harness (DSH) plugin designed to address the development pain points caused by API peak and off-peak pricing strategies. By adding a “Peak-Valley Scheduling” button to the toolbar, it allows users to defer tasks to off-peak pricing periods (such as weekends, holidays, or specific working hours) for automatic execution, thereby reducing costs. Its core features include automatically checking the current pricing period, scheduling tasks with one click, and managing pending tasks in a sidebar. The plugin must be used together with the official `@deepseek-ai/dsh-experimental-schedule-bundle`, with the Bundle properly configured and dependencies installed to ensure data persistence. By encapsulating complex pricing logic, the tool simplifies the workflow and improves development efficiency.
Read Moredsh-jev-context: Context Pruning and Retrieval Based on JEV Relevance Scoring
`dsh-jev-context` is a DeepSeek Harness plugin designed to address context bloat in long conversations and coherence issues when switching topics. It uses the JEV model to assess the relevance of historical fragments to the current request, automatically trims low-relevance content, and restores collapsed information. The plugin provides a visual panel for monitoring token changes, includes fault-tolerance mechanisms, and supports local installation and configuration.
Read Moredsh-output-setting: DeepSeek Harness Output Display and Language Control Plugin
`dsh-output-setting` is a DSH client plugin maintained by developer yijiezhong, designed to address issues in model output presentation and behavior control. It offers three core features through a unified configuration page: node expansion (reasoning blocks and tool calls), code block styling (font size and line height), and output language (auto, Simplified Chinese, and English). Its settings take effect in real time and are written into the system prompt to constrain the language of hidden reasoning channels. The installation command is `plugin_manager(install_bundle, "github:yijiezhong/dsh-output-setting")`. Technical details show that it depends on `schemastery@~3.18.4` and leverages `react` provided by the DSH runtime
Read Moredsh-plugin-git-commit-push: 0-Token Git Commit & Push Plugin
dsh-plugin-git-commit-push is a DeepSeek Harness plugin designed to optimize the Git commit workflow. It supports automatic commit and push with 0 Token (or minimal Token) consumption via slash commands or tool calls. The plugin automatically generates commit messages following the Conventional Commits specification and splits changes into separate commits by type. It is compatible with Windows, macOS, and Linux, requires a DSH restart after installation to take effect, and works only when explicitly triggered by the user.
Read Moredsh-disk-doctor: Disk Space Health Check and Safe Cleanup for DeepSeek Harness
dsh-disk-doctor is a DeepSeek Harness (DSH) storage health-check plugin designed to address limited visibility into DSH disk usage and the risk of accidental deletion. It provides read-only statistics, separately calculates data, programs, and system caches, explains overlaps, and automatically detects multiple paths. Its core feature is a safe “move-only, no-modification” cleanup mechanism that is disabled by default and requires secondary confirmation; it isolates files into a quarantine directory solely by renaming them and supports one-click undo. The plugin has no network access, produces sanitized reports, and supports viewing results sorted by category or Top N. Installation requires DSH version >=0.1.0-rc.2 and Node.js 22+, and currently must be installed via a GitHub repository URL or a .tgz file.
Read Moredsh-workbuddy-connect: Integrate WorkBuddy Desktop App Model into DeepSeek Harness
`dsh-workbuddy-connect` is a DeepSeek Harness (DSH) plugin aimed at simplifying the integration of local desktop application models. It reuses the WorkBuddy desktop app's login session, automatically connects models such as GLM-5.3 and DeepSeek-V4 without additional configuration, and supports image input, credit balance queries, and multi-version coexistence. Installation is simple, but note potential path and compatibility issues on Windows, WSL, or specific TUI versions. This plugin provides a lightweight, practical solution for using desktop models within DSH.
Read MoreDSH Trae Connect: Integrating Trae (SOLO CN) Subscription Models into DeepSeek Harness
DeepSeek Harness (DSH) provides the `dsh-trae-connect` plugin, designed to address the inability to access Trae desktop subscription models via an API. By leveraging a loopback shim to perform protocol conversion between Trae and DSH, the plugin exposes Trae models to DSH through an OpenAI-compatible interface, with support for streaming responses and security isolation. Installation requires DSH 0.1.5–0.1.7, Node.js 22.19.0 or later, and a logged-in Trae environment, and requires exporting plaintext credential snapshots on the host side. This solution is intended for personal learning and research only, depends on a specific environment, and requires caution regarding upstream interface changes and account risks.
Read Moredsh-qoder-connect: Integrating Qoder CN Subscription into DeepSeek Harness
The DeepSeek Harness (DSH) plugin `dsh-qoder-connect` resolves the issue of Qoder CN subscription models not being directly callable within DSH. By reading PAT credentials, the plugin integrates Qoder models into DSH through an OpenAI-compatible interface, enabling provider registration, credential flow, and private protocol adaptation. Users must generate and configure a PAT on the official website, then install the plugin via command. The plugin supports selecting the channel `qoder1` in DSH. Note that this plugin is intended solely for personal learning and research, and users must comply with the terms of service. It may become unavailable if the private protocol is invalidated. The DSH core version must match the 0.1.5–0.1.7 alpha line, and the plugin depends on Node.js ^2.
Read Moredsh-web-search-session: Routing DSH Web Search with Session Models and a Self-Hosted Retrieval Backend
The DSH plugin `dsh-web-search-session` addresses the issues that official web search is billed per token and cannot be attributed to a session ledger. By using a self-hosted retrieval backend (such as SearXNG) to fetch web sources, it forces web search to be billed according to the model selected for the current session. After installation, the plugin changes the `searchProvider` of the `web` entry to `session`. It supports session routing, credential isolation, and visual configuration, and provides a controllable fallback mechanism. The plugin requires Node.js >= 20 and DSH >= 0.1.7-rc.2.
Read MoreDeepSeek Harness Nutstore Backup Plugin
The `dsh-nutstore-backup` plugin for DeepSeek Harness (DSH) incrementally backs up DSH's session records, configurations, and workspace memories to Nutstore via the WebDAV protocol. After installation, it requires logging in with an app password and supports one-click restoration. It features zero dependencies, Chinese path compatibility, and tool integration, while backing up only core data to keep the backup size manageable. Users can back up and restore data through the settings page or by directly invoking model tools, effectively addressing pain points in data migration and long-term preservation.
Read MoreDeepSeek Harness Plugin: dsh-save-chat — Local Markdown Chat Memory Library
`dsh-save-chat` is a DeepSeek Harness plugin designed to address the issue of losing conversation turns after a session ends. It saves conversations as local Markdown files, provides a two-tier storage structure (project-level collections and a personal knowledge base), and offers model retrieval interfaces with support for full-text search and image preservation. The plugin is dependency-free and database-free, storing data only locally. Users can save conversations with a heart button and add them to a knowledge base for on-demand retrieval by the model, making it suitable for developers who need localized knowledge management.
Read More