modlens

Eyes for text-only models.

The first vision plugin for dsh: paste in an image and get structured JSON evidence back, with failover across 10 engines.

2,789 star v3.18.3 Vision / multimodal GitHub

Install

Pick one of three: an already signed-in Claude Code/Codex/OpenCode/Pi (reusing its credentials), a free Gemini API key (Google AI Studio), or the free Antigravity CLI. Known pitfalls: the install command deliberately pins @3.18.3 rather than @latest, because pnpm 11 withholds versions released within the last 24 hours; vision models are excluded automatically, and the vision wrapper is applied only to text-only models; models that cannot be identified are left as they are, to preserve their native image-attachment behavior.

A dsh plugin: install it into your profile, then restart dsh web.

DeepSeek Harness (version pinned on purpose, see pitfalls)

npx -y @deepseek-ai/dsh plugin --profile web add @liustack/modlens@3.18.3

LIUSTACK skills track

npx -y skills add liustack/vibemaster -g

Natural-language install: send this line to your AI

Install and configure the modlens skill following https://github.com/liustack/modlens/blob/main/INSTALL.md, then run the health check

Full tutorial

01Send the install instruction from INSTALL.md straight to your agent and let it install itself and run the health check

02If the health check comes back empty, set up one free engine (a Gemini API key is recommended: 3 minutes, no credit card)

03Once installed, just chat: paste an image or drop in a path, ask anything, and the skill triggers automatically (or go through the "(modlens vision)" entry in the model picker)

config

modlens config set openai.baseUrl https://dashscope.aliyuncs.com/compatible-mode/v1 modlens config set openai.apiKey <key> modlens config set openai.model qwen3-vl-plus # 其他常用项 modlens config set provider <name> modlens config set proxy <url> modlens config set reuse.codex true modlens config set reuse.opencode true modlens config set reuse.pi true modlens config set reuse.grok true

Features

·A 10-engine vision failover chain (6 built-in providers + 4 reusable local agent CLIs)

·Zero-configuration reuse of existing multimodal credentials: no hooks, no wrapper, no local proxy process, and no changes to any harness configuration

·The native modlens_read_image tool returns structured JSON with a full text transcription, reading-order layout regions, and lists of entities and relations

Voices

X @hqmank · 2026-08-05

Gave DeepSeek vision with one skill. npx -y skills add liustack/modlens Repo link 👇

411 likes Source