ModLens: The First Vision DSH plugin for DeepSeek Harness, Giving Text-Only Models Sight
liustack/modlens
ModLens gives text-only models like DeepSeek Harness vision, turning pasted images into structured JSON evidence.
ModLens adds vision to text-only models in DeepSeek Harness (like DeepSeek, GLM). Paste an image and get structured JSON evidence (OCR, layout, semantics) without saving a file first.
dsh plugin --profile web add github:liustack/modlens- Category
- Tools & Capabilities
- Platform
- DSH-Plugin
- Author
- liustack
- Distribution
- Plugin
modlens Key Features
modlens Repository
modlens Repo Summary
ModLens is the first vision plugin for DeepSeek Harness (DSH), maintained by liustack under the MIT license. It solves the problem that text-only models like DeepSeek and GLM cannot read images: users can paste an image directly into the chat without saving it to a file first. ModLens converts the image into structured JSON evidence (including OCR, layout, and semantic information) via a native modlens_read_image tool, enabling the model to understand visual content. The plugin automatically discovers all provider routes carrying text-only DeepSeek or GLM models and adds a wrapped entry per route (e.g., DeepSeek-V4-Flash (modlens vision)), while excluding vision models to preserve native paste behavior.
What core features does the ModLens DSH plugin support?
ModLens supports two paste modes: direct paste (image lands as a private temp file, path enters the composer) or selecting a (modlens vision) model entry (thumbnail stays visible, converted to structured evidence at request time). It auto-discovers and wraps all text-only model routes, remembers user choices, and provides configuration docs, output contract, security docs, and troubleshooting guides for customization and reliability.
How to install ModLens?
Installing ModLens requires a single command using DSH's plugin management. Updating is the same command again. The version is pinned explicitly rather than using @latest to avoid pnpm 11's delay on releases published within the last 24 hours.
dsh plugin --profile web add github:liustack/modlens
Which DeepSeek Harness versions and platforms are compatible with ModLens?
ModLens is designed specifically for DeepSeek Harness and supports all provider routes carrying text-only DeepSeek or GLM models. It automatically excludes vision models to ensure compatibility. The plugin is built with TypeScript and requires Node.js. Last updated in August 2026, it is actively maintained by liustack, and the community can report issues via GitHub Issues.
What are ModLens's configuration and output contract?
ModLens provides detailed configuration documentation (skills/modlens/references/configure.md) and an output contract (docs/output-schema.md) defining the structured JSON evidence format, including OCR, layout, and semantic fields. Security documentation (docs/security.md) explains image handling safeguards. Users can adjust settings to fit specific use cases.
What community and support channels are available for ModLens?
Community support is primarily through GitHub Issues, where users can open issues anytime. Author liustack is active on X (Twitter), where new releases and community updates are announced first. Additionally, a companion ModSearch web tool offers extra visual search capabilities.