Introduction¶
In DSH WebUI, dragging and dropping images is usually sent as a base64 attachment. If the current session uses a text-only model, the image attachment might not enter the conversation. cesaryike/dsh-image-to-path is a DSH plugin designed to change this flow: it takes over the drag-and-drop/paste action for images, saves them to the current session workspace, and inserts the file path as text into the input box, allowing text-only models to access references that can be passed to tools. The DSH plugin ecosystem emphasizes extensibility, and this plugin is maintained by community members under the MIT license.
What is this¶
This plugin targets DSH Web Profile and solves the issue where “text-only models cannot directly process image attachments.” It does not alter the native image channel for multimodal models but instead provides a path-based solution for text-only models. The maintainer is cesaryike. The README also lists DeepSeek AI (DSH agent) for participation in core engineering implementation and fixes.
Core Features¶
- Supports dragging or pasting images and saving them to the current session workspace.
- Inserts file path text into the input box; the model can pass the path to any tool for processing.
- Automatically detects current model capabilities: models supporting images use the native attachment channel; text-only models use the path solution.
- Content-addressed deduplication to avoid re-writing the same image; images already in the workspace are directly referenced in place.
- The upload process includes image magic number validation, server-side naming, atomic writes, a 10 MiB limit, and same-origin checks, without remote requests.
Installation and Activation¶
When installing from GitHub, use the following command:
dsh plugin --profile web add github:cesaryike/dsh-image-to-path
Restart the Web Profile for changes to take effect after installation.
Typical Usage¶
- Drag or paste an image in DSH WebUI.
- The image will be saved to the
dsh-image-to-path-uploads/directory of the current session workspace. - A filename chip appears in the input box; when sent, the model receives the full path.
- The upload directory can be manually cleaned up at any time without affecting plugin operation.
Configuration¶
Configuration can be overridden via the profile’s cordis.patch.yml with the line id image-to-path.
| Field | Default | Description |
|---|---|---|
subdir |
dsh-image-to-path-uploads |
Subdirectory to save |
maxImageBytes |
10485760 |
Single image limit (10 MiB) |
interceptUnknownModels |
false |
Whether models without declared capabilities are intercepted as text-only |
Regarding interceptUnknownModels: Default is off. When enabled, models without declared capabilities are intercepted as text-only; however, it still fails-open and logs a warning if the provider is not registered or the capability query fails.
Use Cases and Notes¶
Suitable for scenarios where text-only models are used in DSH Web Profile but you still want to pass images to subsequent tools in the form of paths. If the current model itself supports images, the plugin will use the native attachment channel and will not affect existing image capabilities.
It is recommended to check the source code and MIT license before installation. After installation, the plugin runs with DSH Web Profile and processes dragged/pasted image data with the permissions of the current dsh process.
Links¶
Repository URL: cesaryike/dsh-image-to-path
The address given in the directory page clue is skillhub.cn Plugin Page. This address comes from the plugin clue; please verify against the actual site content before visiting.