whitelonng/dsh-plugin-describe-image
51Last commit Aug 15, 2026
dsh-plugin-describe-image DSH plugin
This plugin introduces the `describe_image` tool to DeepSeek Harness, enabling text-only models like DeepSeek V4 to understand images. It accepts an image from a local file path, HTTP(s) URL, or attachment reference, sends it to a vision-language model (VLM) at an OpenAI-compatible endpoint (e.g., Qwen-VL, GPT-4o), and returns only the description text. The image never enters the session log.
How to install the dsh-plugin-describe-image DSH plugin
dsh plugin --profile web add github:whitelonng/dsh-plugin-describe-imageCopying does not run this command. Review the repository and version before installing the dsh-plugin-describe-image DSH plugin.
dsh-plugin-describe-image DSH plugin data source
dsh-plugin-describe-image DSH plugin snapshot date: Aug 16, 2026
discovered
What the dsh-plugin-describe-image DSH plugin can do
- Accepts image input as local path, http(s) URL, or JSON attachment reference resolved via the harness attachment service.
- Live configuration card in Settings → Plugins for editing baseURL, model, and API key without restart.
- Per-call API key resolution: inline config → credential seam (`apiKeyEnv`, default `VISION_API_KEY`) → launch environment.
- Security measures: redirects refused, maxBytes/maxOutputTokens/timeoutMs bounds, magic-byte media-type gate, bounded error excerpts, secrets never logged.
- Companion harness changes (shipped in the harness repo) flatten image blocks into `[image attachment …]` notes and accept image prompts on text-only routes.
Where the dsh-plugin-describe-image DSH plugin fits
- Let a text-only model like DeepSeek V4 describe the content of an image during a conversation.
- Integrate image understanding into agent workflows without exposing the raw image to the model.
- Use a remote VLM (e.g., Qwen-VL, GLM-4V, GPT-4o, or local Ollama) as the description backend.
Who the dsh-plugin-describe-image DSH plugin is for
- Users of DeepSeek Harness who want to add vision capabilities to their text-only model.
- Developers building AI workflows that require image description without multimodal model support.
dsh-plugin-describe-image DSH plugin limitations
- Requires an external vision-language model endpoint (OpenAI-compatible) to be configured, e.g., Qwen-VL, GPT-4o, or Ollama.
- Relies on the DeepSeek Harness attachment service to resolve attachment references; standalone usage is limited.
- Plugin installation requires a restart of the application to take effect.
- Only supports single-image input per call; batch processing is not mentioned.
dsh-plugin-describe-image DSH plugin: from the repository README
Quoted from the whitelonng/dsh-plugin-describe-image README, the upstream source of the dsh-plugin-describe-image DSH plugin. Copyright remains with the original authors.
English | [中文](README.zh.md) **DeepSeek Harness 图片理解插件** — a vision-language `describe_image` tool that gives a **text-only model** (DeepSeek V4 and friends) the ability to understand images. A [DeepSeek Harness](https://github.com/deepseek-ai/deepseek-harness) plugin: the model-facing `describe_image` tool. It loads one image — a local file path, an http(s) URL, or a durable attachment reference — and asks a **vision-language model (VLM)** at an **OpenAI-compatible endpoint** (Qwen-VL, GLM-4V, GPT-4o, or a local Ollama endpoint) to describe it. Only the returned **text** crosses into the conversation; the image itself never enters the session log. Keywords: DeepSeek Harness plugin, describe_image tool, image understanding, image description, multimodal, vision-language model, VLM, text-only model, Qwen-VL, GLM-4V, GPT-4o, Ollama. ## Install ```sh dsh plugin --profile web add github:whitelonng/dsh-plugin-describe-image ``` The desktop app's plugin list accepts the same spec in its install box (`github:whitelonng/dsh-plugin-describe-image`); the plugin loads after an application restart. ## Features - **Three input forms**: local path, http(s) URL, or the JSON of an `[image a
Read the full READMERepository license: MIT
dsh-plugin-describe-image DSH plugin questions
How do I install the describe-image plugin?
Run `dsh plugin --profile web add github:whitelonng/dsh-plugin-describe-image` in your terminal, or paste the same spec (github:whitelonng/dsh-plugin-describe-image) into the desktop app's plugin install box. After installation, restart the application for the plugin to load.
Which vision-language models are supported?
Any OpenAI-compatible vision endpoint works. Examples include Qwen-VL (set baseURL to https://dashscope.aliyuncs.com/compatible-mode/v1), GLM-4V, GPT-4o, or a local Ollama endpoint. Configure the endpoint in Settings → Plugins → 'Image understanding'.
Does the image itself get stored in the conversation history?
No. The image is loaded, checked, and sent only to the configured vision-language model. The session log and the text-only model see only the returned description text. The plugin never writes the image into the conversation.
How do I configure the API key for the vision model?
The plugin resolves the API key in three layers, in order: an inline `apiKey` in the config, the credential seam (`apiKeyEnv`, default `VISION_API_KEY`), and then the launch environment. The key is never written to logs. You can set it via the plugin's configuration card or environment variable.
Is the plugin safe against malicious images?
Yes. The plugin refuses redirects, gates media types by magic bytes, enforces bounds on image size, output tokens, and timeout, truncates error excerpts, and never logs secrets. These measures prevent hostile images or endpoints from exfiltrating data.