Directory entry
Anionex
agent-vision-toolkit plugin: Vision Tools for Text-Only LLMs
Agent Vision Toolkit provides vision tools and skills for text-only LLMs, enabling image Q&A, OCR, UI restoration, and GUI automation, with optional integration for several agents.
What this plugin does
The agent-vision-toolkit plugin is a vision toolkit and skill designed for text-only LLMs. It enables image Q&A, long-screenshot OCR, frontend UI restoration, and GUI automation. It offers optional seamless integration for Codex, Claude Code, Pi, Oh My Pi, and OpenCode, allowing text-only models to handle visual tasks effectively.
The agent-vision-toolkit plugin provides vision tool CLIs and a skill that teaches agents when to use each tool. It supports pasted images and built-in image tools. The toolkit includes intent-aware image Q&A, grounding, detection, tracing, cropping, pixel diff, long-screenshot OCR, foreground extraction, dominant-color analysis, and HTML screenshots. It offers a transparent local proxy and single-file native plugins for seamless integration. The project has been verified in real Codex and DeepSeek sessions.
Documented capabilities
Intent-Aware Image Q&A
The agent-vision-toolkit plugin captures the user's or model's latest intent when viewing an image, producing details needed for the current turn instead of a broad description.
Long-Screenshot OCR
It can extract long screenshots, chat histories, and scrolling pages by finding low-content cut bands, OCR each chunk in order, preserving speakers and timestamps, and merging overlaps.
UI Restoration and GUI Automation
The plugin can rebuild a UI from a screenshot or design, restore icons and graphics, and operate a GUI from screenshots by locating controls and verifying state changes.
Practical use cases
Rebuild a UI from a Screenshot
Using the agent-vision-toolkit plugin, an agent can reuse project components and assets, combine code-native UI, extracted visuals, and visual comparison to align a page or component.
Extract Long Screenshots
The plugin can process long screenshots and chat histories by OCRing chunks in order, preserving speakers and timestamps, and merging duplicated overlaps.
Operate a GUI from Screenshots
An agent can locate a control, perform an action, capture the screen again, and verify the resulting state before continuing, enabling GUI automation.
Review before you install
Read the source, recent releases, requested permissions, network access, and maintainer history before trusting a third-party plugin.
Frequently asked questions
What is the agent-vision-toolkit plugin?
The agent-vision-toolkit plugin is a vision toolkit and skill for text-only LLMs, providing image Q&A, OCR, UI restoration, and GUI automation, with optional integration for several agents.
Which agents can integrate with the agent-vision-toolkit?
It offers optional seamless integration for Codex, Claude Code, Pi, Oh My Pi, and OpenCode. It also supports DeepSeek Harness as a native Profile Bundle.
How does the agent-vision-toolkit handle images?
It supports both pasted images and built-in image tools. The included skill teaches the agent what to inspect, which tool to choose, and how to verify results.