dsh-vision-toggle: Add a Per-Model Image Input Toggle for DeepSeek Harness
The `dsh-vision-toggle` plugin provides a graphical solution for DeepSeek Harness (dsh), eliminating the need to manually edit `settings.yaml` to enable a model's image input. It allows the vision input toggle to be managed independently for each model, loads routed models and applies changes immediately, and supports Chinese and English. After installation, you must restart the application, and it is only applicable to models under the `llm-pi-ai` route.
Read Moredsh-plugin-effort-declare: Declaring Reasoning Effort Levels and Image Input for Manually Added OpenAI-Compatible Models
`dsh-plugin-effort-declare` is a DeepSeek Harness (DSH) plugin intended to solve the problem where manually added third-party models cannot declare `reasoningEfforts` (reasoning effort levels) and `input` (image input) capabilities. By adding a page under Settings, it writes these capabilities to the official namespace, thereby restoring the Effort selector and image upload functionality in the UI. The plugin is released under the MIT license and applies only to manually added models using the OpenAI protocol. Core features include configuring reasoning effort levels per model, declaring image input support, providing protocol dialect presets, and exposing version information. Installation requires DSH 0.1.2-alpha.2 or later; after installing via a terminal command, navigate to “Settings → Reasoning”
Read Moredsh-luna-vision-bridge: Add Native Image Input Support for DSH
`dsh-luna-vision-bridge` is a host-side DSH LLM plugin designed to overcome the image input limitations of the native input box. It registers a new `luna-vision-bridge` provider to intercept the `addImages` and `submit` workflows: user-uploaded images are transcribed into text via Codex Luna, and the resulting text is passed to a configured downstream plain-text model. The plugin supports native interactions such as pasting and drag-and-drop, provides a web-based configuration interface, but relies on the Codex CLI in the host environment.
Read More