Otherdsh-plugin
dsh-vision-bridge
A DSH plugin that enables text-only models like DeepSeek to accept images by automatically translating image content into text via an external vision model (Alibaba Cloud Bailian qwen-vl). Works transparently without switching models or manual commands; pure server-side, compatible with read_image tool.
Stars
1
Forks
0
Open issues
0
Last push
Aug 16, 2026
Latest release
—
24h Growth
+0
7d Growth
+0
Installation
No reliable install command was detected. Check the project README for installation instructions.
Overview
A DSH plugin that enables text-only models like DeepSeek to accept images by automatically translating image content into text via an external vision model (Alibaba Cloud Bailian qwen-vl). Works transparently without switching models or manual commands; pure server-side, compatible with read_image tool.
- Automatically converts image messages to text descriptions for text-only models
- Wraps resolveModelInfo to declare image support for selected providers
- Wraps adapter.stream to intercept and translate image blocks before model serialization
- Pure server-side; no GUI or session history modifications
- Does not interfere with models that natively support images
- Configurable via environment variables; supports caching and image compression
- Optional reuse of claude-vision-skill configuration via VISION_SKILL_DIR
- Verified with DSH 0.1.0-rc.6; MIT licensed