DeepSeek Harness Vision Bridge Plugin
`dsh-vision-bridge` is a vision plugin designed for DeepSeek Harness (DSH), aiming to compensate for the limited image-processing capabilities of text-only models. The plugin supports configuration through a GUI or YAML, allowing users to choose a DSH provider or a custom OpenAI-compatible API. Its core features include: enabling models to proactively invoke the `vision_bridge_read_image` tool for OCR and semantic analysis; automatically uploading images pasted into the input field and generating reference markers; and providing connectivity testing and visual configuration. It is compatible with DSH 0.1.1+ and enables the flexible construction of vision agents by reusing existing channels or using standalone configuration.
Read MoreDeepSeek Harness (DSH) Multimodal Vision Bridge Plugin
The DSH (DeepSeek Harness) plugin `dsh-vision-bridge` is designed to provide visual capabilities to text-only models that do not natively support images, such as DeepSeek. The plugin proxies `llm/stream` requests, converts images into structured Chinese descriptions that are injected into the context, and provides two tools: `view_image` and `ocr_image`. It supports native multimodal routing detection to optimize performance and allows hot-updating configuration from the settings page. After installation, users can upload images directly in a conversation; the plugin handles them automatically or via tool calls, without requiring any modifications to the DSH core source code.
Read Moredsh-vision-bridge: Enabling Text-Only Models to Receive Images
`dsh-vision-bridge` is a DeepSeek Harness plugin designed to address the inability of text-only models to process images. It leverages the `qwen2.5vl` model run locally via Ollama, converting images into text descriptions sent to the DeepSeek API to enable localized visual processing. The plugin provides a native image experience, transparent request-layer adaptation, structured output, and coordinate support. It includes nine vision tools, such as description, OCR, and region queries, and supports caching of duplicate images. Once installed, the plugin registers a “twin” route and injects a tool set, with all operations completed locally to ensure privacy and security.
Read More@zzdream67/dsh-vision-bridge: Enabling Pure-Text Models to Read Images in DeepSeek Harness
This plugin intercepts the LLM pipeline and replaces images with transcripts generated by a vision model before requests reach the upstream, enabling text-only models to support image inputs. Installation requires pnpm. Core features include image interception, model declaration injection, and Web-based visual configuration. Users can manage vision model declarations from the settings page by testing them, directly enabling them, or reloading them. The plugin also supports automatic declaration management, cached transcription results, and timeout settings, and it handles restoration of declaration state upon uninstallation.
Read Moredsh-vision-bridge: General-Purpose Vision Bridge for DeepSeek Harness
`dsh-vision-bridge` is a runtime intermediary plugin for DeepSeek Harness (DSH), designed to give text-only chat models native image-processing capabilities. The plugin intercepts and rewrites image blocks, converting them into descriptive textual context for the text model, while also supporting native pass-through and approximately 40 vision tools (such as OCR, VQA, and structured extraction). It provides three operating modes: hybrid (default, automatically describes images and preserves tool usage), llm (pure rewriting), and tools (explicit tool invocation only). Users can install it via the terminal and configure it in settings. Note that the Web UI must be restarted after installation, and a Node.js 20+ environment is required.
Read Moredsh-vision-bridge: Web DSH DeepSeek Session Image Bridge
dsh-vision-bridge is an MIT-licensed plugin designed specifically for the web version of DSH, aiming to address the issue where pure text models like DeepSeek cannot directly process image inputs. By automatically delegating images to an external vision model (Alibaba Cloud Bailian qwen-vl) for recognition and converting the results into text to pass to the current model, this plugin enables pure text models to understand image content. Core features include allowing image uploads, automatic image-to-text recognition, providing a qwen-vl model fallback chain, supporting text-to-image generation (generate_image), and reusing recognition result caches. It is applicable only to the web version of DSH; for the desktop version, the built-in plugin dsh-tool-vision must be used. Installation requires enabling the plugin in the DSH profile and configuring environment variables such as DASHSCOPE_API_KEY. If ds... is enabled simultaneously...
Read Moredsh-vision-bridge: Accessing DSH session images via non-visual main models
dsh-vision-bridge is a DeepSeek Harness (DSH) plugin designed to address the issue where main models without visual capabilities (such as deepseek-v4-pro) cannot process user images. By invoking external visual interfaces, this plugin automatically converts images into text descriptions during the model request assembly phase. The main model only receives text, thereby preserving image thumbnails in user messages and achieving content understanding. The plugin provides a view_image tool that supports local paths, URLs, and data URLs. It does not include a built-in vision service provider and requires explicit configuration of interfaces and keys for OpenAI, xAI Grok, etc. Installation supports npm commands or a one-click script from the GitHub repository (install.ps1), and manual configuration is also supported. The core process includes copying files and registering the plugin.
Read Moredsh-vision-bridge: Add a description bridging layer for DeepSeek V4 / V4-Flash image messages
dsh-vision-bridge is a third-party image processing plugin for DSH, designed to address the limitation where text-only models, such as DeepSeek V4/V4-Flash, cannot natively process image context. By invoking an OpenAI-compatible Vision Language Model (VLM) to automatically convert user input images into text descriptions before passing them to DeepSeek, the plugin achieves image understanding without altering the original text processing workflow. Core features include: automatically recognizing and converting image content; dynamically declaring model support for image input; intercepting and replacing image blocks before requests; allowing models that natively support images to pass through directly; caching descriptions based on attachmentId to prevent double billing for repeated sends; and graceful degradation without interrupting the conversation when recognition fails. Installation requires the dsh CLI and pnpm as prerequisites.
Read MoreDSH-Vision-Bridge: Adding Image Understanding to DeepSeek Harness's Text-First Model
DSH-Vision-Bridge is an MIT-licensed plugin for DeepSeek Harness, designed to provide image understanding capabilities to text-first models. This plugin bridges image requests to external vision APIs (supporting Gemini, OpenAI, and Anthropic-compatible interfaces) and returns only the analysis results in text form, preventing image blocks from being directly forwarded to upstream models. Its core advantage lies in preserving the original model list and selection experience, with independent management of vision bridge preferences through the "glasses" control next to models, without interfering with model selection logic. The plugin supports configuring multiple isolated vision providers in settings, strictly limiting usage to HTTPS endpoints, and keeping sensitive information such as API keys on the Host side. Installation requires compatibility with Harness 0.1.x and Node.js 2.
Read More@dsh-extension/dsh-vision-bridge: On-Demand Visual Capability Integration for Text-Only DSH Sessions
`@dsh-extension/dsh-vision-bridge` is a third-party plugin that provides on-demand visual capabilities for text-only DSH sessions. It addresses the issues where text models cannot process image blocks and sending long history directly to vision models is prohibitively expensive. Mechanically, the plugin preserves the original images in the session and UI, only reformatting the images into text tokens at the text model's input layer. When the model needs to understand an image, it invokes the `vision_describe` tool, sending only the specified images (1-4) and a focused question to an OpenAI-compatible vision endpoint, thereby avoiding the transmission of lengthy session histories. The plugin supports installation via npm, with configuration options including enable status, base URL, API key (with environment variable support), and vision model name. Configuration priority follows the order: settings.
Read More