CatchMe captures and organizes a user's entire digital footprint to make AI agents truly personal. It provides a lightweight, privacy-first solution for CLI agents, enabling them to access a comprehensive, searchable memory of user activities like coding, research, and file management. This enhances AI interactions with deep, contextual personalization.

How It Works

CatchMe records user activities (windows, keyboard, mouse, files) and structures them into a Hierarchical Activity Tree, enhanced by LLM-generated summaries. Its novel approach bypasses vector embeddings, using LLM-driven tree traversal for retrieval. This allows for efficient, top-down navigation of structured memory, enabling complex reasoning without the overhead of traditional vector databases.

Quick Start & Requirements

Install via git clone, conda create -n catchme python=3.11, and pip install -e .. Requires macOS Accessibility/Input Monitoring/Screen Recording permissions or Windows Administrator privileges. Initialize with catchme init for LLM setup. Run recording with catchme awake, visualize/chat via catchme web, or query with catchme ask -- "question". An LLM with multimodal support is required.

Highlighted Details

• Always-On Event Capture: Comprehensive, event-driven recording of user actions and context.
• Intelligent Memory Hierarchy: Auto-organizes activity into a multi-tier tree with LLM summaries.
• Tree-Based Retrieval: Novel vectorless retrieval using LLM tree traversal.
• Zero-Config Agent Integration: Drop-in skill file for seamless CLI agent integration.
• Ultralight & Privacy-First: Minimal RAM (~0.2GB), local storage, offline LLM support.
• Rich Web Interface: Visual exploration and natural language chat.

Maintenance & Community

The project encourages community contributions and is inspired by related open-source projects. Specific details on maintainers, sponsorships, or dedicated community channels are not provided.

Licensing & Compatibility

The README does not explicitly state a software license. This is a critical omission for due diligence, leaving the terms of use, modification, and distribution unclear. Compatibility focuses on CLI agents, with strong support for local LLM providers.

Limitations & Caveats

CatchMe requires a configured LLM and specific OS permissions for full functionality. Using cloud LLM APIs for summarization may pose privacy risks. The absence of a stated license is a significant adoption blocker.