DamonKoy/dsh-web-ui#dsh-tool-describe-image
Gives a text-only model image understanding via a vision-language model, exposed as a describe_image tool.
Gives text-only models vision: when a conversation mentions an image (local path, http(s) URL, or session attachment), the describe_image tool sends it to a configured OpenAI-compatible vision endpoint and returns the answer. Only the returned text enters the conversation — the image itself never enters the session log. Since text-only models have no image entry in the input box, the plugin adds an image button that inserts an attachment reference into your draft; the tool also accepts a prompt argument for precise custom instructions (OCR, UI diagnosis, translation). Endpoint, model, key and default instruction are configured live under Settings > Plugin config.
Install
dsh plugin --profile web add @linxin666/dsh-tool-describe-imagenpm package @linxin666/dsh-tool-describe-image 0.3.6 (registry-verified 2026-08-27) from the dsh-web-ui family. Restart dsh web; configure endpoint/model/key under Settings > Plugin config > 'Image understanding'. Ported from whitelonng/dsh-plugin-describe-image (upstream deepseek-harness packages/vision/tool-describe-image), Apache-2.0.
Compatibility
DeepSeek Harness web profile. Requires a configured OpenAI-compatible vision endpoint (Qwen-VL, GLM-4V, GPT-4o, local Ollama endpoint...).
Details
- Repo: DamonKoy/dsh-web-ui#dsh-tool-describe-image
- Category: Tools & Capabilities
- Stars: 6
- Version: npm @linxin666/dsh-tool-describe-image 0.3.6 (registry-verified 2026-08-27)
- Last push: 2026-08-15
- First seen: 2026-08-15
Recent updates
describe_image tool; image button in composer; prompt argument for OCR/UI diagnosis; privacy-preserving (image never enters session log); live config.
FAQ
- Does the image enter the conversation?
- No — only the returned text enters the conversation; the image itself never enters the session log.
- Which vision models work?
- Any OpenAI-compatible vision endpoint: Qwen-VL, GLM-4V, GPT-4o, or a local Ollama endpoint — configured live in settings.
- Can I give custom instructions?
- Yes — the tool accepts a prompt argument (OCR, UI diagnosis, translation) that beats a generic description.
Alternatives
whitelonng/dsh-plugin-describe-image · 121103qwq/dsh-vision-sidecar · Einskyle/dsh-llm-vision-bridge