llm-deepseek-image-admit

DeepSeek official models declare only the text input modality, so the DSH apiproxy rejects any image message before it reaches the agent loop. Even with an image-bridge or similar transcription plugin installed, the image never gets past the admission pre-check.

This plugin takes over the DeepSeek official provider route and replicates the exact apply logic (same config surface, settings integration, credential resolution). The sole difference: resolveModel now returns the image modality alongside text. The model selector stays unchanged and no manual provider switching is required, making it fully transparent to the user. It imports the already-built official artifact directly, with zero extra build dependencies.

Best for developers who use DeepSeek models inside DSH and want to send images that are later transcribed to text by a pre-step plugin such as image-bridge. Note that this plugin only admits images and does not transcribe them; without a pre-step transcription plugin, raw image blocks will still be rejected by the DeepSeek API.