Catchme
"CatchMe: Make Your AI Agents Truly Personal"
Last verified:
What is Catchme?
CatchMe is an open-source personal memory layer that captures your entire digital footprint and makes AI agents truly personal. It transforms raw digital activity into structured, searchable memory through three concurrent stages: Capture, Index, and Retrieve. Six background recorders silently track window focus, keystrokes, mouse movement, screenshots, clipboard, and notifications, then auto-organize them into a Hierarchical Activity Tree (Day → Session → App → Location → Action) with LLM-generated summaries at each level.
The tool uses tree-based retrieval instead of traditional vector search, allowing the LLM to traverse your memory tree top-down, select relevant nodes, inspect raw data like screenshots or keystrokes, and synthesize precise answers. This enables complex cross-day reasoning and precise evidence gathering from raw activity history without the complexity of vector embeddings and databases.
CatchMe is designed for developers, AI agent builders, researchers, and anyone who wants their AI assistants to have persistent personal context. It ships as an agent-compatible skill for CLI agents like OpenClaw, NanoBot, Claude Code, and Cursor. Key use cases include personal coding assistants (recall what you were coding), personal deep research (track what you were reading), personal files manager (track file changes), and digital life overview (track app usage and workflows).
Catchme pricing
Pricing model: Freemium
CatchMe is completely free and open-source under the Apache-2.0 license. There are no paid plans or subscription tiers. The only cost is for LLM API usage if you choose cloud providers: approximately $0.42 per 2 hours of intensive use with qwen-3.5-plus via Aliyun DashScope, or ~$5.00 per 2 hours with gemini-3-flash-preview via OpenRouter. You can run it fully free using local LLMs (Ollama, vLLM, LM Studio) with no API costs. Disk usage is ~200MB and runtime RAM is ~0.2GB.
Catchme pros
- Open-source and free to use with Apache-2.0 license
- 100% local storage - all data stays on your machine in ~/data/
- No vector embeddings or vector databases required
- Ultralight footprint - only ~0.2GB runtime RAM usage
- Offline-first support with Ollama, vLLM, LM Studio
- Six comprehensive background recorders for complete Capture
- Hierarchical Activity Tree with 5-tier organization
- LLM-generated summaries at every tree level
- Top-down tree-based retrieval for precise answers
- One-file agent integration for instant setup
- Rich interactive web dashboard at port 8765
- Natural language chat with your digital footprint
- Event-driven recording with no timer delays
- Cross-day reasoning capabilities
- Supports 18+ LLM providers including local options
- Interactive timelines and memory tree navigation
- Real-time system monitoring in web interface
- CLI commands for agent memory queries
- SQLite + FTS5 efficient storage backend
- Multimodal support for text and images
Catchme cons
- Requires an LLM for background summarization and retrieval
- Cloud API usage may expose private activity data
- macOS requires Accessibility, Input Monitoring, Screen Recording permissions
- Windows requires Administrator privileges for global input monitoring
- LLM costs can accumulate with intensive use (~$0.42-$5.00 per 2 hours)
- No official release yet - still in development on main branch
- Only 405 stars - relatively new and smaller community
- No multi-device recording yet - single machine only
- Full retrieval takes 5-20 seconds per query
- Multimodal LLM required - text-only models won't work
Frequently asked questions about Catchme
What is CatchMe?
CatchMe is an always-on personal memory layer that captures your entire digital footprint and makes AI agents truly personal. It records window focus, keystrokes, mouse movement, screenshots, clipboard, and notifications, then organizes them into a hierarchical Activity Tree with LLM-generated summaries. Agents can query this memory via CLI commands to understand your past activities, code sessions, research, and workflows.
Is CatchMe free?
Yes, CatchMe is completely free and open-source under the Apache-2.0 license. There are no paid plans. The only potential cost is LLM API usage if you use cloud providers. You can run it entirely free using local LLMs like Ollama, vLLM, or LM Studio with no API costs.
How do I install CatchMe?
Clone the repository and install with: git clone https://github.com/HKUDS/catchme.git && cd catchme, then create a conda environment: conda create -n catchme python=3.11 -y && conda activate catchme, and finally install: pip install -e . On macOS, grant Accessibility, Input Monitoring, and Screen Recording permissions. On Windows, run as Administrator.
What LLM providers does CatchMe support?
CatchMe supports 18+ providers including OpenRouter, AiHubMix, SiliconFlow, OpenAI, Anthropic, DeepSeek, Gemini, Groq, Mistral, Moonshot/Kimi, MiniMax, Zhipu AI, DashScope/Qwen, VolcEngine, BytePlus, and local options like Ollama, vLLM, and LM Studio. Any OpenAI-compatible endpoint works by setting api_url and api_key directly.
How does CatchMe's retrieval work?
CatchMe uses tree-based retrieval instead of vector search. The LLM traverses your Activity Tree top-down, reads summaries at each level (Day → Session → App → Location → Action), selects relevant branches, and drills down to inspect raw evidence like screenshots and keystrokes. It then synthesizes a precise answer. This enables complex cross-day reasoning without vector embeddings.
Is my data private with CatchMe?
Yes, 100% of your raw data (screenshots, keystrokes, activity trees) stays locally in ~/data/ and never leaves your machine. You can run fully offline with local LLMs (Ollama, vLLM, LM Studio). However, if you use cloud APIs, they will be used to summarize your daily activities, so untrusted endpoints may expose private data.
Which AI agents can integrate with CatchMe?
CatchMe ships as an agent-compatible skill for CLI agents including OpenClaw, NanoBot, Claude Code, and Cursor. Your agents query memories via CLI commands only. You can use the light skill (agent queries via CLI) or full skill (agent manages full CatchMe lifecycle autonomously) by copying CATCHME-light.md or CATCHME-full.md to your agent's skills folder.
What is the Hierarchical Activity Tree?
The Activity Tree is CatchMe's memory core with 5 tiers: Day → Session → App → Location → Action. Raw events auto-organize into this structure, and each node gets LLM-generated summaries. This provides structured, multi-level views of your digital life - you can browse high-level summaries or dive into granular details without vector complexity.
What are CatchMe's upcoming features?
The roadmap includes: Multi-Device Recording (capture GUI activities across all machines via LAN synchronization), Dynamic Clustering (adaptive clustering algorithms reflecting actual work patterns to reduce costs), and Enhanced Data Utilization (unlock deeper insights from screenshots and metadata beyond current pipelines). Star the repo to follow future updates.