Modlens
First vision plugin for DeepSeek Harness. Paste an image, get structured JSON: OCR, layout, semantics.
Install
Forwards to pnpm inside the web profile. Confirm the README if the author documents a .dsh-plugin path.
dsh plugin --profile web add github:liustack/modlens
What it is
Modlens is the vision bridge most people reach for when DeepSeek Harness is running a text-only model. You paste a screenshot or photo; the plugin returns structured evidence — OCR text, layout blocks, and semantic labels — instead of a fuzzy caption. It also markets itself as a vision bridge for other text-only coding agents (Claude Code, Codex, Pi, OpenClaw), so the same skill can travel. If your job is “read this UI” or “extract the error from this screenshot,” start here.
Source: liustack/modlens. Official harness: deepseek-ai/deepseek-harness.
Listed from public GitHub topic dsh-plugin, snapshot 2026-08-13.
Related in Vision
Vision toolkit for text-only LLMs. Works with Codex, Claude Code, Pi, OpenCode — and tags dsh-plugin.
Harness-native vision toolkit: image Q&A, long-screenshot OCR, UI restoration, grounding, pixel diff, Artifacts.