Introduction to the DeepSeek Harness Plugin: meng-114/dsh-image-tiler
The DeepSeek Harness (DSH) plugin `meng-114/dsh-image-tiler` is designed to address the issue of detail loss caused by large image inputs. It segments a large image into coordinate-tagged tiles and a global overview image, performs tiling via the `tile_image` tool, and uses `read_tiles` to support reading strategies such as region-based tile selection, semantic localization, and workbench checkbox selection. The plugin supports multiple formats, requires DSH version 0.1.7 or later, and strictly restricts paths to the current workspace, helping vision models retain details while reducing token consumption and improving the accuracy and efficiency of processing complex images.
Read Morexby-wild-animal-detection: Wild Animal Detection Plugin
The DeepSeek Harness plugin `xby-wild-animal-detection` encapsulates wild animal visual recognition capabilities, aiming to reduce the integration cost for developers. It provides three input methods (image URL, BASE64 encoding, and file path) to identify animals and supports setting the API key via the `set_xby_apikey` command. Once installed, it can be invoked in DSH workflows to analyze wild animal bounding boxes and confidence scores in images. Before use, users must register on the official website to obtain an API key, and the plugin is released under the MIT License.
Read Moredsh-vision-pro-bridge: Completing the Vision Capabilities of DeepSeek-V4-Pro
The `dsh-vision-pro-bridge` plugin adds vision capabilities to the `deepseek-v4-pro` model in DeepSeek Harness (DSH). Through a routing proxy, it leverages the official DeepSeek API to first use `deepseek-v4-flash-vision-exp` to transcribe images into text, then passes the text and paths to the Pro model for processing. The plugin features zero third-party dependencies and content-addressed persistence. After installation, users can select the “DeepSeek-V4-Pro (Vision Bridge)” model to perform screenshot analysis while retaining the Pro model’s coding capabilities, making it suitable for use cases such as layout analysis and OCR.
Read Moredsh-screen-reader: Giving Agents Eyes
The `dsh-screen-reader` plugin is designed to address Agent screen-perception automation and supports the Windows platform. It provides two core tools: `see_screen`, which directly captures the screen and sends the image to the model, and `see_diff`, which locates changed regions using local pixel differencing. The plugin enforces image normalization, decouples visual tasks, and allows the model to focus on "what happened" rather than "whether anything changed." Installation must be done via a profile bundle. This plugin is suitable for Agents to self-verify their operations, but it has limitations, including low measurement precision, no handling of temporal causality, Windows-only support, and an attachment library that can grow without bound.
Read MoreDeepSeek Harness Host-Level Visual Plugin: dsh-visibridge
`dsh-visibridge` is a host-level vision plugin for DeepSeek Harness that compensates for the limited image-processing capabilities of text-only models. Through the `analyze_image` and `capture_image` tools, it sends images to local Ollama or cloud APIs and returns structured JSON evidence containing OCR results, layout analysis, and other details. It supports local processing as well as cloud models such as Xiaomi MiMo and DeepSeek. After installation, users only need to place images in the workspace or capture them with a camera, and the AI can reason based on the evidence. The plugin supports multi-backend switching, secure API key redaction, and intranet access controls, providing a localized and secure solution for document processing and vision-feedback agents.
Read Moredsh-vision-subagent: Enabling Vision Capabilities for DeepSeek Harness Text Agents
The `dsh-vision-subagent` plugin provides visual analysis capabilities to the main conversation by delegating image processing to independent subagents running on routes such as MiniMax or Kimi. It implements context isolation, multi-turn visual reasoning, and cost separation, while supporting image pasting on the web and high-fidelity reproduction. Installation requires route configuration. Currently, it is limited to local file reading and does not support remote URLs, aiming to address context overflow issues.
Read Moredsh-plugin-vision: DeepSeek Harness Vision Assistant Plugin
The `dsh-plugin-vision` plugin for DeepSeek Harness (DSH) addresses the lack of vision capabilities in the official API by registering the `vision_describe` tool. It enables the main model to call any OpenAI-compatible vision API to describe images. Installation must be performed using the official script, and the plugin supports configuration of custom API endpoints, API keys, and caching strategies. It relies on the global dependency tree, so the `dsh plugin add` command must not be used during installation. Additionally, when used with DSH Desktop, it can automatically convert pasted images to file paths, provided that dependencies are installed correctly.
Read Moredsh-mingmu: Equipping DeepSeek Harness's Text-Only Model with External "Eyes"
dsh-mingmu is an ESM plugin designed specifically for DeepSeek Harness (dsh), aiming to enable pure text models to handle visual content. When a model is detected as unable to recognize images, the plugin automatically invokes a built-in vision model to process the image and return a text response, preventing workflow interruption. It supports capability-based triggering, automatic upgrade strategies, visual configuration, and secure API key storage, addressing image-processing limitations for pure text models in the dsh environment.
Read Morejolly-dsh-vision: DeepSeek Harness Vision Bridge Plugin
`jolly-dsh-vision` is a visual bridging plugin designed for DeepSeek Harness (DSH), enabling text-only large models to process pixel data. It introduces a dedicated vision model as the "eyes" and provides two visual pathways: invoking the `vision` tool to obtain structured evidence (such as OCR and layout), and enabling the `(ds vision)` twin model to directly support image input. Installation requires the DSH CLI and configuration of the brain and eye models. The plugin includes security features such as zero-touch API key handling and SSRF protection, but has limitations including limited format support and reliance on a single route. The project is released under the MIT License, has zero code dependencies, and aims to improve the accuracy of visual reasoning with a minimal architecture.
Read MoreIntroduction to the DeepSeek Harness Image Understanding Plugin
`dsh-vision-analysis` is an image understanding plugin for DeepSeek Harness, designed to address the lack of visual capabilities in text-only models. It supports vision endpoints in OpenAI or Anthropic format, allowing integration with models such as Qwen or GPT-4o as well as local Ollama, and provides eight analysis modes, including OCR and chart data extraction. It is easy to install and accepts images as local files, URLs, or Base64 inputs. With image bridging configured, agents can paste images directly into conversations for analysis. The plugin prioritizes privacy by processing images internally only and returns structured JSON data for OCR and chart data extraction.
Read Moredsh-vision-resident: DeepSeek Harness Resident Vision Subagent
`dsh-vision-resident` is a resident vision plugin for DeepSeek Harness, designed to address the limitation of purely text-based main models that cannot process images directly. By mounting hooks, it enables the main model to “see” images without launching independent subagents, significantly reducing overhead. The plugin supports zero-subagent reasoning, persistent conversational memory, session-isolated history, and durable storage. Users can read local images through tools, allow the main model to automatically interpret images, and view recognition history via the interface. Installation requires manually mounting the code, offers flexible configuration, and supports custom vision models. Despite known limitations such as model-switching constraints, it provides a lightweight and efficient solution for scenarios with frequent visual tasks.
Read Moredsh-plugin-vision-toolkit: Integrating Vision Capabilities into DeepSeek Harness Agents
This plugin equips DeepSeek Harness (DSH) agents with visual perception capabilities. By registering system skills, it converts images into text descriptions or coordinate information, addressing the limitation that agents cannot process images directly. It provides four core tools: glance (description/question/OCR), ground (element localization), detect (instance detection), and crop (cropping). Installation requires specifying a configuration file and setting the visual API key and model. The tool processes data as text without accessing raw pixels, aiming to extend the application boundaries of agents.
Read Moredsh-vision-plugin: Empowering DeepSeek Harness Text Models with Vision Capabilities
The `dsh-vision-plugin` is a plugin for DeepSeek Harness (DSH) that enables visual analysis capabilities for text-only models, such as DeepSeek v4. It uses an intermediate layer to send images to a vision model, such as Qwen3.7-plus, for transcription into text, which is then processed by the main model. The plugin supports pasting images directly into a conversation to ask questions and provides the `vision_analyze` tool for analyzing local files. Before use, image input support for the model must be declared in the configuration file. The plugin is compatible with DSH v0.1.0-rc.8 and later.
Read Moremimo-vision: DSH Native Vision Plugin
mimo-vision is a native plugin designed for DeepSeek Harness (DSH), providing text description capabilities for host models that lack native vision support. As a first-class citizen of DSH, it registers the `describe_image` tool and uses DSH infrastructure to manage dependencies and credentials, sending images to a multimodal model and parsing the resulting descriptions. The plugin supports routing to free or paid APIs, preferring free APIs by default, and can be installed with the command `dsh plugin --profile web add mimo-vision`. It requires Node.js `^22.19` or `>=24`, and an OpenCode API Key must be configured. When used, the host model can invoke the tool to generate text descriptions for images at specified paths.
Read Moredsh-vision-api-localorweb: Integrate a Local Vision Model API and Equip the DSH Assistant with a Pair of “Local Eyes”
DeepSeek Harness (DSH) does not support image recognition by default. This plugin enables access to local models to read image content while preserving privacy. It supports analyzing images through local OpenAI-compatible endpoints, such as LM Studio or Ollama, keeping all data on the local machine. Users can configure the endpoint address, model name, and API key in the settings interface without modifying configuration files. After installation, simply provide an image path in the conversation to obtain a text description. Note that direct image sending may be limited; using the file path is recommended, and the local service must be compatible with the standard interface.
Read MoreDeepSeek VisionPlus: Text goes to DeepSeek, images automatically routed to free vision model pools
When using DeepSeek Harness (DSH), there is an unavoidable limitation: DeepSeek's official models are text-only and cannot process images. When screenshots, charts, or interface photos appear in a message, one must either manually transcribe the content or switch the entire conversation to another multimodal model, resulting in a break in context. DeepSeek VisionPlus takes a different approach: text requests continue to be handled by the DeepSeek official API, while images are automatically routed to a pool of free visual models, with the understanding results returned to the same conversation. DSH's philosophy is that "everything is a plugin," and this capability is implemented as a plugin. Below is an introduction to its features, installation, and daily usage.
Read Moredsh-vision-suite: A Windows Plugin to Supplement DeepSeek Harness with Vision Capabilities
# Foreword When building agents with DeepSeek Harness (DSH), relying on DeepSeek's text model for reasoning is fine, but many tasks require "seeing": reading error messages from screenshots, comparing layout differences between versions, or extracting primary colors from design drafts. The common practice is to manually send the image to another multimodal service and then transfer the results back to the conversation, breaking the workflow in the middle. dsh-vision-suite addresses exactly this issue. While retaining the official `deepseek-official` text routing, it integrates image understanding, OCR, screenshot cropping, pixel comparison, primary color extraction, and secure webpage screenshots into Harness. Text reasoning remains handled by DeepSeek; the plugin will only forward...
Read Moredsh-img: Image understanding and local visual tools for DeepSeek Harness
dsh-img is a community visual plugin for DeepSeek Harness designed to provide image understanding capabilities to pure text models. By integrating with a vision backend or local pixel tools, it enables `analyze_image` for image Q&A, OCR, layout understanding, and `vision_ground` for element localization. The plugin supports Zhipu GLM-4V-Flash, Tongyi Qwen-VL, Ollama local models, and OpenAI-compatible endpoints, and includes key-free local tools such as `vision_crop` and `vision_ocr`. Installation requires Node.js >= 20 and the DeepSeek Harness core, and it is installed into a specific profile via `pnpm add`. When using an API backend, the API key must be injected.
Read Moredsh-unknown-theme: Port the DeepSeek official website visuals to DeepSeek Harness
dsh-unknown-theme is a community plugin maintained by ycqaq233 under the MIT license, designed to optimize the visual effects of the DeepSeek Harness (DSH) Web workspace to closely resemble the official look of deepseek.com. Not an official component, this plugin is implemented using pure CSS/DOM/Canvas, has zero dependencies, requires no build process, and supports adaptive light/dark dual themes. Its core features include porting the official site's WebGL fluid background, mouse-interactive grid (HeroGrid), particle fish-shaped logo animation, title spotlight effects, and UI component styles featuring a frosted glass texture. Installation is recommended by executing `dsh plugin --profile web add ycqaq233/dsh` via the git repository.
Read Moredsh-tool-describe-image: Add image understanding capability to DSH
`sala003/dsh-tool-describe-image` is an MIT-licensed plugin within the DSH ecosystem designed to equip text models with image processing capabilities. By connecting to any OpenAI-compatible vision endpoint (such as Bailian, Zhipu AI, etc.), it converts images into text descriptions or structured HTML for use with DeepSeek. Core features include a Web interface for "paste and recognize," tool invocation via path to describe images, querying DeepSeek account balance, and providing a customizable entry point for a "Whale Girl" desk pet. The plugin supports zero-configuration startup, allowing API keys to be configured later via environment variables or the settings panel. Installation requires DSH 0.1.0-rc.6+, installed globally via npm and enabled. It supports PNG, JPG, WebP, and GIF formats, with a single-image limit of 8Mi.
Read Moredsh-vision: A plugin to add visual capabilities to DSH
`dsh-vision` is a native DSH plugin maintained by `xiaoshihou514`, aiming to supplement visual capabilities for DeepSeek Harness. The plugin features a minimalist design, with core functions including adding visual support, utilizing free Zhipu interfaces, and employing a local Qwen-VL 2B visual small model. Installation requires deploying the DSH core first (requiring Node engine ^22.19.0 || >=24 and specific peer dependencies), and the plugin is released under the MIT license. Please note that the installation command provided in the README references the incorrect plugin address `dsh-weixin#main`; the command should not be executed directly to install `dsh-vision` until the maintainer corrects it. The plugin is suitable...
Read MoreDS-Vision-Plugin: A Web Image-to-Text Plugin for DeepSeek Harness
ds-vision-plugin is an open-source plugin designed for DeepSeek Harness, aimed at addressing the issue that text-based models cannot directly process image inputs. The plugin automatically converts pasted or dragged images into text within the web interface for DeepSeek models to continue reasoning, eliminating the need for users to manually switch tools or models. Its core features include support for a four-model racing mechanism (such as the agnes and glm series) to achieve fast and reliable first valid responses; routing support for Baidu OCR or local Tesseract, with fallback to a vision-language model when unavailable; and the ability to integrate custom OpenAI-compatible models or local runtimes (Ollama/LM Studio). Additionally, the plugin offers comprehensive diagnostic tools and YAML hot-reload functionality.
Read Moredsh-visual-plugin: Enhancing Image and Video Understanding for DSH Web UI
dsh-visual-plugin is a visual media plugin developed by jyh20030112 for the DeepSeek Harness Web interface. Images are processed through the DSH native attachment system and the current visual model pathway, with a right-side panel recording reproducible description histories. Videos undergo validation via FFprobe, H.264 normalization, and keyframe extraction using PySceneDetect before being submitted to the model for analysis. The installation command is `dsh plugin --profile web add dsh-visual-plugin`, with video functionality requiring the host machine to have FFmpeg >= 6.1 and PySceneDetect 0.7.x pre-installed. Licensed under MIT, with zero runtime npm dependencies.
Read Moredsh-media-skills: Enhancing DeepSeek Harness with Free Image Recognition and Generation
dsh-media-skills is a DSH plugin maintained by MJorgin (MIT, v0.3.0), supporting rc.7/rc.8/v0.1.1-rc.x. It offers automatic captioning for stickers, vision-review for image reading, and media-tools for image generation. The vision routing includes GLM-4V-Flash and DeepSeek-V4-Flash-Vision-Exp, while image generation utilizes SenseNova U1 Fast and Kolors. API keys are not stored in the repository, and the installation command is `dsh plugin --profile <name> add github:MJorgin/dsh-media-skills`.
Read MoreModLens: Augmenting Pure Text Models with Visual Capabilities
ModLens is a DSH vision plugin maintained by liustack (MIT, v3.25.0). After pasting an image, an independent vision engine generates structured JSON evidence for OCR, layout, and semantics, which can be used by pure text models like DeepSeek and GLM for answering. It supports both direct pasting and the (modlens vision) model paths, with six built-in providers and the ability to reuse local CLI login states. DSH installation command: npx -y @deepseek-ai/dsh plugin --profile web add @liustack/modlens@3.25.0.
Read MoreGive Pure Text DeepSeek Harness Agent Eyes with dsh-vision-router
dsh-vision-router is a DeepSeek Harness vision routing plugin maintained by ysr666: the vision model serves only as the eyes, while DeepSeek continues reasoning. It features a built-in key-free OVH vision link and includes 14 pixel-level tools (Q&A, localization, cropping, pixel comparison, OCR, etc.). With a one-command installation, you can click "👁 Vision" before sending an image to perform multi-step image recognition just like using regular tools. This article verifies the installation command, tool list, and usage notes based on the SkillHub directory page and GitHub README.
Read MoreEquip Plain-text DeepSeek Harness Agent with "Eyes" Using dsh-vision-router
### 版本1（正式学术/技术文档译法，适配海外开发者语境）： DSH-Vision-Router is a DSH tool and capability plugin (MIT License, v1.4.4) maintained by ysr666. It treats vision models as its "eyes" and DeepSeek as its "brain", and provides free-key-free anonymous fallback service from OVH by default, along with 11 pixel-level tools including question answering, positioning, cropping, pixel comparison and OCR. This article verifies the installation commands, usage of automatic image recognition model groups and permission precautions in accordance with the community directory and GitHub README. --- ### 版本2（更贴合国内开源项目对外宣传的自然译法）： Maintained by ysr666, DSH-Vision-Router is a DSH utility and capability plugin under MIT License (v1.4.4). It takes vision models as its "eyes" and DeepSeek as its "brain", offers free anonymous fallback service powered by OVH by default, and has built-in 11 pixel-level tools such as Q&A, positioning, cropping, pixel comparison and OCR. This article checks the installation commands, usage of the automatic image recognition model suite and permission notes against the community repository and GitHub README.
Read MoreAdding Visual Capabilities to DeepSeek Harness Plain Text Model Using dsh-vision-toolkit
DSH Vision Toolkit is a DeepSeek Harness interface enhancement plugin maintained by Anionex, which connects agent-vision-toolkit to Web/Headless Profile. After pasting images with a plain-text model, users can switch to the Vision Toolkit variant to complete intent-based image question answering, element positioning, long-screenshot OCR and UI restoration with pixel comparison. The directory installation command is `dsh plugin add github:Anionex/dsh-vision-toolkit`, it is licensed under MIT, and the current npm version is 0.1.20.
Read MoreAdd Vision Capabilities to the Plain-text Models of DeepSeek Harness with ModLens
ModLens is a DeepSeek Harness visual plugin maintained by Liustack (MIT License, version 3.18.1 at the time of writing). When plain-text DeepSeek/GLM cannot recognize screenshots, it registers the `modlens_read_image` function to convert pasted images into JSON containing OCR results, layout information and semantic data. This article verifies the installation command, two image pasting paths, engine configuration and output contract according to the community directory and GitHub.
Read More