A persistent visual memory system for AI agents to store, analyze, and recall images, charts, and website screenshots.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install minds-eye
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install minds-eye using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Multimodal Memory?
Multimodal Memory is a sophisticated skill designed to bridge the gap between text-based interactions and visual context. It allows AI agents to maintain a long-term visual archive, ensuring that diagrams, screenshots, and UI mockups shared in past sessions remain accessible and searchable. By leveraging this tool within the Openclaw Skills ecosystem, developers can build agents that truly see and remember history across multiple sessions.
This skill uses a dedicated analysis pipeline to process images via GPT-4o, bypassing the vision limitations of primary models. Whether you are tracking UI changes over time or recalling a specific chart from last week, this skill provides the persistent storage and retrieval layer necessary for advanced multimodal workflows.
Multimodal Memory Use Cases
- Saving and tagging user-uploaded images, photos, or technical diagrams for future reference.
- Capturing and remembering website UIs or specific web page layouts via URL screenshots.
- Recalling past visual context during a conversation by searching through stored descriptions.
- Finding similar past content by analyzing a new query image and matching it against the visual database.
How Multimodal Memory Works
- The agent detects an image path or a URL request in the conversation context.
- It executes a capture or analysis script which processes the visual data using GPT-4o.
- Metadata, including descriptions and tags, is automatically extracted and stored in a local SQLite database.
- A human-readable Markdown summary is updated to provide a quick overview of recent visual memories.
- The agent performs semantic or keyword searches against this database to retrieve relevant visual history when prompted.
Multimodal Memory Setup
Ensure you have Python 3 installed. If you plan to capture website screenshots, install the Playwright dependencies within the Openclaw Skills environment:
pip install playwright && python -m playwright install chromium
The skill automatically creates its storage directory at ~/.multimodal-memory/ upon the first execution.
Multimodal Memory Data Schema & Taxonomy
The skill organizes information within ~/.multimodal-memory/ using the following structure:
| Component | Description |
|---|---|
images/ |
A directory containing local copies of all captured and analyzed images. |
metadata.db |
An SQLite database storing image paths, GPT-4o descriptions, source URLs, and tags. |
memory.md |
A human-readable Markdown file that provides a summarized log of stored visual content. |
scripts/ |
Python utilities for analysis (analyze.py), capture (capture_url.py), and search (search.py). |
Multimodal Memory Advanced Features
- Automated website UI tracking with the integrated Playwright screenshot engine.
- Cross-session visual persistence allowing agents to reference images from weeks ago using Openclaw Skills.
- Multi-source support, distinguishing between charts, diagrams, website captures, and general images.
- Flexible tagging system using the
--extra-tagsargument for customized metadata organization. - Seamless integration with the Openclaw Skills framework to provide vision-less models with robust visual memory capabilities.
SKILL.md
Loading
Related Openclaw Skills
emo-img
A comprehensive utility for searching, managing, and sending sticker and emoji images across various chat platforms.
Sequential Read
An AI-powered reading assistant that processes long-form text sequentially to simulate evolving human perspective and deep thematic synthesis.
Wiki.js CLI Skill
A comprehensive command-line interface for managing Wiki.js instances through the GraphQL API.
TTS WhatsApp Voice Messaging
An AI-powered skill to automate high-quality text-to-speech voice message delivery on WhatsApp in over 40 languages.
Quant Trader Daily
A professional workflow and task management system for quantitative and stock traders to automate reporting and risk monitoring.
Agent Bounty Scanner
A precision discovery engine that scores and ranks agentic bounties based on budget, urgency, and capability alignment.
Featured*
