Einskyle/dsh-llm-vision-bridge

Native LLM-provider vision bridge: images pasted in the chat are described by a vision model (Qwen3-VL via pi-ai/llama.cpp) and the text description is fed to text-only DeepSeek for the reply — image admission, routing and compaction all run through harness-native mechanisms, with an LRU description cache and 503 retry.

A vision bridge for DeepSeek Harness that routes image analysis to a vision-capable model when the primary model cannot accept images, so the agent keeps working on multimodal tasks. Install from GitHub (plain JS, no allowBuilds) or npm; manual install path documented for environments without pnpm.

Vision & Multimodal ★ 4 updated 2026-08-14
View on GitHub ↗

Install

dsh plugin --profile web add github:Einskyle/dsh-llm-vision-bridge

Two documented routes: dsh plugin --profile web add github:Einskyle/dsh-llm-vision-bridge (plain JS, no build step, no allowBuilds needed) or the npm route dsh plugin --profile web add dsh-llm-vision-bridge (npm package dsh-llm-vision-bridge 0.1.2, registry-verified 2026-08-25). Then restart the web service: pnpm dsh web. Manual install without pnpm: copy the package into %USERPROFILE%\.dsh\profiles\web\node_modules\dsh-llm-vision-bridge\ and edit the profile's package.json.

Compatibility

DeepSeek Harness Web profile. Vision fallback for text-first model routes — the check is only a fallback; to actually send images set the main model to deepseek-vision.

Details

Recent updates

v0.1.2: GitHub + npm install routes; restart via pnpm dsh web; manual copy-based install documented for pnpm-less setups.

FAQ

How do I install it?
Either dsh plugin --profile web add github:Einskyle/dsh-llm-vision-bridge or dsh plugin --profile web add dsh-llm-vision-bridge (npm). Then restart the web service with pnpm dsh web.
Is a build step needed for the GitHub install?
No — it is plain JS with no build step, so a GitHub install needs no allowBuilds permission.
When does the vision bridge engage?
It is only a fallback: image analysis is routed to a vision-capable path when the primary model cannot accept images. To actually send images, set the main model to deepseek-vision.

Alternatives

GXX182/dsh-vision-bridge · Sorwcyra/ds-vision-plugin · oil-oil/dsh-vision

More plugins in Vision & Multimodal

Browse more in Vision & Multimodal

Guides for Vision & Multimodal plugins