dsh-image-router: Enabling Text-Only Models to “See” Images
The DeepSeek Harness plugin `dsh-image-router` addresses the inability of text-only models to process image inputs. Before prompt admission, it leverages a vision model to automatically convert images into text descriptions, while keeping the session model selection unchanged. Users can paste images into a conversation for automatic text conversion or use the `describe_image` tool to analyze local images. The plugin supports configuring an existing vision model or a custom endpoint, and features comprehensive audit logs, test coverage, and a graphical configuration interface. It requires DSH 0.1.5-rc.1 and Node.js ≥ 20.
Read Moremeow-vision: Giving Text Models a Pair of Eyes and Multimodal Models a Pair of “UI Rendering” Eyes
meow-vision is a DeepSeek Harness (DSH) plugin designed to address the issues that text models cannot read images and multimodal models lack real UI rendering screenshots. It includes two core tools: `meow_vision` provides visual capabilities to text-only models by configuring a multimodal model to recognize images, while `meow_preview` allows multimodal models to view the actual rendering of UI components through dev server screenshots. The plugin supports configuring the Vue demo path and can automatically manage file writes and dev server startup. It must be installed via the DSH plugin manager, supports only the `web` profile, and performs no telemetry or remote reporting at runtime.
Read Moredsh-auto-vision: Equip DeepSeek Harness's text-only main model with vision tools
## Introduction When developing agents with DeepSeek Harness (DSH), you are likely to encounter this issue: the built-in `read_image` tool directly injects image chunks into the current model's context, and it is only usable if the current main model supports images. If your main model is a pure text model such as `deepseek v4 flash`, calls to `read_image` will be rejected—the main model cannot perceive the image, effectively breaking the image reading path. dsh-auto-vision adopts a different approach: it forwards images internally to a multimodal model for recognition, ensuring the main model only processes text throughout the process. This article explains its underlying principles, installation, and configuration. ## What is dsh-auto-vision
Read Moredsh-vision-helper: Persistent Vision Plugin for DeepSeek Harness
dsh-vision-helper is a zero-dependency persistent vision plugin built for DSH, designed to integrate image analysis capabilities non-intrusively. It registers the `vision_analyze` tool, supports local path or data: URI inputs, and returns analysis text via a configured multimodal model. The plugin allows customization of provider, model, temperature, max output tokens, and image dimensions, with configurations taking effect immediately without requiring a restart and featuring a dedicated Web settings interface. Its core advantages include zero additional dependencies, global availability, persistent storage, and robust mechanisms such as automatic handling of invisible characters and deduplication. Installation supports npm, GitHub, or manual copying, and requires model routes or compatible endpoints to be pre-configured within llm-pi-ai. Users can configure it via the Web settings page or JS...
Read More