dsh-analyze-image-tool
Run the following command in DeepSeek Harness:
dsh plugin install CaseyTso/dsh-analyze-image-tool
Paste the following prompt into your AI chat to install this plugin:
Run dsh plugin install CaseyTso/dsh-analyze-image-tool in your terminal to install; the full source code is available at https://github.com/CaseyTso/dsh-analyze-image-tool
About this plugin
Most workhorse models in DeepSeek Harness (deepseek-v4-flash, for instance) are text-only and cannot natively accept images. Worse, DSH's API gateway rejects any image-plus-text-only-model combination before plugin hooks even fire, so the vision path is hard-blocked at the infrastructure level.
dsh-analyze-image-tool bridges vision into text-only sessions. It registers an analyze_image tool that any model can invoke; the plugin then forwards the image as base64 to any OpenAI-compatible vision or multimodal endpoint and returns a structured response (text, model ID, token usage) back to the model. No vendor is hard-coded: a single baseURL + apiKey + model triple covers SiliconFlow, Alibaba DashScope compatible mode, Zhipu, OpenRouter, a local Ollama instance, or OpenAI itself. For pasted images, a built-in runtime bridge intercepts the incoming prompt when the active model is text-only, converts image blocks into readable attachment references, and guides the model to call analyze_image to actually see the picture. Vision-capable models pass through untouched. A small eye-icon panel in the WebUI session header lets you edit endpoints, keys, prompt templates, save multiple presets, and run a live connectivity test with a real 64x64 test image all without restarting the server.
Who is this for? DSH users running text-only models who regularly need the model to read a screenshot, a chart, or a photographed document. No model swap, no extra proxy, no vendor lock-in. Install the plugin, point it at any vision endpoint you already have, and pasting an image into the conversation just works.
Use Cases
- Paste a screenshot into a text-only model session and ask the model to read a chart or diagram
- Switch between SiliconFlow, DashScope, or a local Ollama endpoint in one click from the WebUI panel
- Run fully offline image analysis with local Ollama and Qwen3-VL on scanned documents or handwritten notes
Best For
- Developers running text-only DSH models who still need image understanding
- Users who want to pick any vision provider without vendor lock-in
- Teams that require offline image analysis and prefer not to send images to the cloud
Related Plugins
ModLens is a vision plugin for DeepSeek Harness that gives text-only models sight by reading images pasted directly into chat, with zero-config setup and multiple vision engines.
On-demand vision for text-only DeepSeek Harness agents: built-in free keyless vision chain and 14 vision tools, routing image turns as tool calls to vision models with pixel fidelity, no Python needed, one-command install.
Give text-only models in DeepSeek Harness eyes, enabling image Q&A, long-screenshot OCR, UI restoration, and GUI visual tasks.
An armor-breaking plugin for DeepSeek: rewrites instruction priority, expands refusal-blocked phrases, persists across sessions, and shows a green active indicator.