wulusai2333/mimo-vision

describe_image tool: a vision bridge that sends images to mimo-v2.5 through the opencode Zen API (credential OPENCODE_GO_API_KEY, free route first with paid fallback) and returns text descriptions for text-only models, with native passthrough and ImageMagick transcoding of SVG/TIFF/HEIC formats.

mimo-vision is a native vision bridge for DSH: it registers a single describe_image tool that sends an image to a mimo-v2.5-series multimodal model and returns a text description to the main model — built for text-only main models like deepseek-v4-flash. PNG/JPEG/GIF/WebP/BMP are sent natively; SVG/TIFF/HEIC/PSD/ICO/EXR/JP2/JXL/AVIF are auto-transcoded to PNG via ImageMagick when installed (downscaled to ≤2048px long edge to save tokens); other formats return an explicit error, never a silent send. Image read goes through ctx.fs with a 20 MiB cap; relative paths resolve against the session workspace.

Vision & Multimodal ★ 0 updated 2026-08-16
View on GitHub ↗

Install

dsh plugin --profile web add mimo-vision

npm mimo-vision 0.1.0 verified 2026-09-04 (repository field → github.com/wulusai2333/mimo-vision; README EN primary w/ zh edition). Install: dsh plugin --profile web add mimo-vision (npm) or dsh plugin --profile web add github:wulusai2333/mimo-vision (source, prebuilt lib/ — no build approval needed). Remove with dsh plugin --profile web remove mimo-vision (by package name, not the github: spec).

Compatibility

DeepSeek Harness (web profile); key resolved via ctx.credentials from OPENCODE_GO_API_KEY → OPENCODE_API_KEY (env > ~/.dsh/.credentials.yaml > .env); ImageMagick optional for SVG/TIFF/HEIC/PSD/ICO/EXR/JP2/JXL/AVIF transcode.

Details

Recent updates

Registration is reversible by construction (apply only registers the tool); uninstalling cleans up cleanly.

FAQ

Which image formats are supported?
PNG/JPEG/GIF/WebP/BMP natively; SVG/TIFF/HEIC/PSD/ICO/EXR/JP2/JXL/AVIF are transcoded to PNG via ImageMagick when installed (≤2048px long edge). Anything else returns an explicit error.
Which model does it use?
A mimo-v2.5-series multimodal model; the API key resolves from OPENCODE_GO_API_KEY then OPENCODE_API_KEY through DSH credential layering.
Is this a standalone service?
No — it is a native DSH plugin: apply() only registers describe_image as a first-class tool, and dependencies/credentials/subprocesses go through DSH seams.

Alternatives

ruby1304/dsh-vision-subagent · good-boy4069/dsh-vision-guard

More plugins in Vision & Multimodal

Browse more in Vision & Multimodal

Guides for Vision & Multimodal plugins