liustack/modlens

ModLens – Vision plugin for DeepSeek Harness

What it is – ModLens is a plug‑in that gives text‑only LLMs (e.g., DeepSeek‑V4‑Flash, GLM‑5.3) the ability to “see”. It lets you paste an image directly into a chat, sends the picture to a configurable vision engine, and returns a structured description that the language model can use in its reply.

Key capabilities

• One‑click install for DeepSeek Harness (dsh) via npx @deepseek-ai/dsh plugin … @liustack/modlens@<version> or via the generic skills.sh command.
• Zero‑config start – re‑uses any existing vision services you already have (Claude Code, Codex, OpenCode, Pi, etc.) and falls back to built‑in providers.
• Six built‑in vision providers (Gemini, OpenAI‑compatible, Anthropic, Antigravity‑CLI, Claude‑CLI, Kimi‑CLI) plus four local CLIs that can be reused.
• Failover chain – if the primary provider fails (auth, rate‑limit, network) ModLens automatically tries the next one, recording each attempt.
• Direct paste – images are processed from a temporary file created behind the scenes; no manual file‑path handling is required.
• Structured output – returns full transcription, layout regions, entity‑relation lists, and confidence notes, so the model can quote exact details rather than hallucinate.
• Lightweight – no local proxy daemon or config hacks; uninstalling is just deleting the plugin folder.

Installation

1. DeepSeek Harness: npx -y @deepseek-ai/dsh plugin --profile web add @liustack/modlens@3.26.6 (or latest version).
2. Other harnesses: npx -y skills add liustack/modlens --skill modlens --global.
3. Run the built‑in health‑check (modlens doctor) and follow the prompts to configure a vision engine (free Gemini key, Antigravity CLI, or any OpenAI‑compatible endpoint).

How to use

• After installation, simply paste an image into the chat window of a text‑only model. The modlens_read_image tool triggers automatically, sends the picture to the selected vision engine, and the model replies with a grounded description.
• Subsequent questions about the same image need no additional paste.
• You can pin a specific provider (modlens config set provider gemini-api) or let the failover chain decide.

Ecosystem & related tools

• ModSearch – sibling plug‑in that adds web‑search capability to models lacking internet access.
• DeepSeek Harness Desktop – a GUI front‑end for the Harness service.
• dsh‑market – plugin marketplace where ModLens is listed alongside 800+ community plugins.

License – MIT (see LICENSE).

Bottom line – ModLens turns any text‑only DeepSeek or compatible LLM into a multimodal assistant with minimal setup, leveraging a flexible set of vision back‑ends and a seamless paste‑into‑chat workflow.