Agent Vision Toolkit

Agent Vision Toolkit is an open-source set of vision tools that lets text-only coding agents understand images, read screenshots and operate GUIs.

Agent Vision Toolkit gives coding agents that only handle text a way to work with images. It passes the agent's reason for looking, a focus hint, to a vision model so the description fits the task, then uses five command-line tools: glance to understand an image, ground to locate a named element, detect to list elements, trace to extract exact SVG shapes and crop to cut out regions. Together they support image questions, OCR of long screenshots with overlap removal, rebuilding a UI as HTML and CSS with pixel checks, GUI automation from screenshots, and turning diagrams into SVG, HTML or Mermaid. The toolkit installs as a CLI and a vision-tools skill and works with Codex, Claude Code, Pi, Oh My Pi and OpenCode. It is open source on GitHub.

Agent Vision Toolkit at a glance

Best for: Software engineer
Runs on: CLI, open source
Pricing: Free and open source
Released: August 2026
Category: AI agent tool
Website: agent-vision.anionex.me

Key features

Focus hints that pass the agent's intent to the vision model
glance, ground, detect, trace and crop tools
Long screenshot OCR with overlap removal
UI recreation as HTML and CSS with pixel checks
GUI automation from screenshots
Works with Codex, Claude Code and OpenCode

Use it for

Let a coding agent rebuild a UI from a screenshot
Read a long chat screenshot accurately
Automate a desktop app with a text-only model

Alternatives to Agent Vision Toolkit

ModLens — Converts screenshots and charts into structured JSON data for coding agents.
Agent Service Toolkit — Complete LangGraph agent service template, from agent to chat UI
Adyen — Share AI agent resources across coding tools and repositories easily.
AGENTS.md — It separates agent-specific instructions from human-focused README files.
Spec Kit — Provides structured processes and templates for AI coding agents.
Agent Toolkit for AWS — Provides official MCP servers and agent skills tailored for AWS services.
Vision Agents — It enables building low-latency AI agents with support for any model.
AgentsView — It keeps your AI coding session archive local by default while tracking costs and activity.
Agentation — It translates clicks into CSS selectors and source paths for agents.
fast-agent — Improves agent accuracy and cost efficiency via MCP integration.
OpenViking — Unifies agent memory, knowledge RAG, and skills into a single directory structure.
CodePilot — Consolidates multiple AI agents and local model connections in one workspace.

All Agent Vision Toolkit alternatives