Skillsdsh-plugin

modlens

ModLens is a vision plugin for DeepSeek Harness and a vision bridge for text-only coding agents. It reads images pasted into chat and returns structured JSON evidence such as OCR, layout, and semantics for DeepSeek, GLM, and other text-only models.

Stars
4,271
Forks
131
Open issues
2
Last push
Oct 15, 2026
Latest release
v3.27.1
24h Growth
+0
7d Growth
+37

Installation

dsh plugin --profile web add @liustack/modlens@3.27.1

Overview

ModLens is a vision plugin for DeepSeek Harness and a vision bridge for text-only coding agents. It reads images pasted into chat and returns structured JSON evidence such as OCR, layout, and semantics for DeepSeek, GLM, and other text-only models.

  • Paste images directly into chat; ModLens reads the image without requiring you to save a file and pass a path first.
  • Returns structured JSON evidence with OCR transcription, reading-order layout regions, and entity/relation lists.
  • Auto-discovers eligible text-only DeepSeek, GLM, or MiMo Pro provider routes and adds a `(modlens vision)` entry per route.
  • Combines six built-in vision providers and four reusable local agent CLIs behind one failover chain.
  • Rotates comma-separated API keys after authentication, rate-limit, or quota failures; other failures keep provider failover.
  • Installs as a DSH plugin with one command or as a skill via skills.sh.
  • Zero-config start can reuse existing Claude Code, Codex, OpenCode, and Pi setups, or use a free Gemini API key.
  • DSH settings include a ModLens card to switch engines and choose which local CLIs `auto` mode may reuse.

Related plugins