liustack/modlens
Vision bridge for text-only models: paste an image, get structured JSON evidence (OCR, layout, semantics).
Install
via npm
Resolves a published tarball rather than cloning the repository, and installs without any extra setup. Swap `web` for your profile name if you run another one.
via GitHub · npm package
Installing from GitHub runs the project's build script, which pnpm blocks until you allowlist it — run the command once and pnpm prints the exact key to add under `allowBuilds` in ~/.dsh/profiles/web/pnpm-workspace.yaml.
What happened when we ran it
Installed cleanly when we ran it
Every command here is run in a throwaway container against a clean profile, and the result is whatever the harness recorded — not a guess from the source. Last run 11d ago.
Show it in your README
For maintainers: the badge serves this listing's latest sandbox verdict, so a re-run updates it on its own — and it links readers to the full result here.
Due diligence
What it is — ModLens bridges text-only models with vision, allowing pasted images to obtain structured JSON evidence including OCR, layout, and semantics.
Who it is for — When using DeepSeek Harness with text-only models like DeepSeek-V4-Flash to analyze pasted images, it is suitable. For users of native multimodal models, this plugin is not necessary because native models retain their built-in image pasting capability.
Watch out — Sandbox testing passed, with successful installation in a fresh profile and registration with Harness. No obvious issues found.
The verdict — I would install it because it outputs structured visual evidence without saving the image to a file path first.
Generated by grok-4.6, and a starting point rather than a verdict. Where it says a plugin installs or does not, that is from a real run in a clean profile — everything else is read off the repository. Trust the source over this.
modlens is a DeepSeek Harness plugin that gives text-only DeepSeek and GLM models an image-reading path. It handles images pasted into the chat or supplied by path, then returns structured JSON evidence such as OCR, reading-order layout regions, entities and relations. The plugin uses the native modlens_read_image tool. In the standard paste flow, the host places the image in a private temporary file and puts its path in the composer. A (modlens vision) model entry instead keeps the thumbnail in the message and converts it to evidence at request time. ModLens discovers provider routes carrying text-only models and creates wrapped entries for them, while leaving confirmed vision models on their native paste path.
It is intended for agents whose selected model cannot inspect images, provided at least one vision engine is reachable. The README describes built-in providers and reuse of existing logins from Claude Code, Codex, OpenCode or Pi; reads can therefore consume an API key, endpoint quota or a reused local CLI login. The terminal surface matters: setup and operation may inspect local harness CLIs and invoke their authenticated routes, and pasted images pass through a temporary local file. It is the wrong choice when the selected model already has native vision support, when image data must not be sent to an external vision engine, or when no configured provider or reusable CLI can read images.
modlens_read_image is the native tool exposed to the text-only model. A pasted image can arrive as a private temporary-file path in the composer, or through a selected (modlens vision) model entry, where the thumbnail remains in the message until request time. The result is structured evidence intended to ground the model’s answer in what the image contains. A path can also be supplied directly in chat.
The plugin auto-discovers provider routes whose metadata positively confirms that they carry text-only DeepSeek or GLM models. It adds a wrapped (modlens vision) entry for each such route. Confirmed native vision models are excluded, and unconfirmed models are left alone. A stock setup is documented as exposing DeepSeek-V4-Flash (modlens vision) and DeepSeek-V4-Pro (modlens vision); additional routes such as opencode-go or zai can receive their own entries.
The README describes six built-in providers and four reusable local agent CLIs. Documented provider examples include gemini-api, which needs a Gemini API key, openai, which takes an OpenAI-compatible key, baseUrl and model, and anthropic, which needs an Anthropic API key. antigravity-cli is also documented as a no-key channel after sign-in. Existing logins in Claude Code, Codex, OpenCode and Pi can be offered for reuse, with each reused read labelled by the quota it used.
The setup checks available local credentials and asks before reusing them. The documented Antigravity sign-in command is agy. The plugin can inspect local harness CLI availability and invoke an approved route, so its terminal and credential access should be reviewed alongside the image data sent to the selected provider.
Written from the project's own documentation and kept in sync with it. Where the two disagree, the source is authoritative — read the README on GitHub
Same category
zhu1090093659
A `describe_image` vision tool for text-only models: images (local path, URL, attachment) go to a configurable OpenAI-compatible vision endpoint and only the returned text enters the session.
Installed cleanly when we ran it
Anionex
为纯文本模型"看图“设计更好的视觉工具箱和技能，支持多图理解，图片问答，前端UI还原、GUI 自动化等，并可选无缝接入多个主流agent，直接识别粘贴图片｜ A vision toolkit and skill designed for text-only llms — image Q&A, long-screenshot OCR, frontend UI restoration, and GUI automation, with optional seamless integration for Codex, Claude Code, Pi, Oh My Pi, and OpenCode
Installed cleanly when we ran it
ysr666
Free vision for text-only agents: built-in keyless vision chain plus pixel tools (Q&A, grounding, crop, pixel diff, colors, OCR, SVG trace, cutout, screenshots); paste an image to use it.
Installed cleanly when we ran it
Anionex
Vision for text-only models: paste an image and the model switches to a Vision Toolkit variant for image Q&A, multi-image comparison, long-screenshot OCR, screenshot-to-UI reproduction, element grounding, and pixel diff. No API key by default — images are processed by the author-hosted free service, 100 per machine per day; configurable to your own provider.
Installed cleanly when we ran it
omdsh-dev
Chrome sidebar extension that lets DeepSeek Harness operate your browser directly, no vision capabilities required. 一款 Chrome 侧边栏扩展程序，可让 DeepSeek Harness 直接操控您的浏览器，无需视觉能力。
Installed cleanly when we ran it
oxbshw
DeepWatch, powered by Watch Skill: a local-first perception and verification stack for AI agents. Understand video, audio and screen activity, keep timestamped evidence, and independently verify what a tool or workflow actually did. MCP, CLI, REST, Web and DeepSeek Harness.
Installed cleanly when we ran it
