Vision for text-only DSH models
nlqh7/dsh-vision-router · model-adapters
Visualize the DeepSeek Harness pure text model: auto-distribution of user configuration visual models with a codex-style chat box drag tow, v4 without switching models.
★ 2GitHub stars
0Forks
2026-08-25Last updated
TypeScriptLanguage
MITLicense
Key features
- Allows a pure text master model (e.g., deepseek-v4-pro) to receive pictures in the chat box: the picture is converted into a text description by route to the user-configured visual model and then back to the main model.
- The plugin does not contain any visual provider address or key.
- When you are loaded and restarted, drag a chat box with a pure main text model and the picture is automatically translated into a text description.
- The main model sees only text and does not need to switch the model.
Requirements
- The front-end hard intercept is still judged by the master model capacity - the plugin has been rewritten with a return value for the call path, so that it needs to actually pass through ctx.llm.resolveModelInfo (both check and main request before the front-end sends the picture here).
- If a future front end changes the ability to search, the picture will still be blocked.
Install command
dsh plugin --profile web add github:nlqh7/dsh-vision-router