✳️ ego_proxy

Run powerful AI language models FREE completely offline. No "Big Brother" with your data, stay 100% local.

Personal Assistant with Memory

Talk naturally with your AI assistant , who remembers everything you've said and can provide context-aware responses and integrate with Google Calendar for event management.

What You Need

Python 3.10 or newer (tested with Python 3.14). A GPU helps but isn't required. Models download automatically from Hugging Face (default is about 1.5GB). That's it.

Getting Started

# Create virtual environment
python3 -m venv .venv
source .venv/bin/activate  # On Windows: .venv\Scripts\activate

# Install
# Run the optimized installation script
./scripts/install.sh

# or
# Install dependencies via pip
pip install --prefer-binary --upgrade-strategy only-if-needed -r requirements.txt

Using It

Personal Assistant:

./assistant.sh

The assistant automatically saves all conversations, extracts metadata (people, topics, sentiment), retrieves relevant context when you ask questions, and integrates with Google Calendar for event management. See ASSISTANT.md for detailed documentation and GOOGLE_CALENDAR_SETUP.md for calendar integration setup.

Features

• Perfect Memory: Never forget a conversation - everything is automatically saved and searchable
• Context-Aware: Retrieves relevant past conversations to provide informed responses
• Metadata Extraction: Automatically identifies people, topics, dates, sentiment, and categories
• Google Calendar Integration: Create events naturally with phrases like "schedule meeting tomorrow at 3pm"
• Semantic Search: Find past conversations using natural language queries
• Timeline View: Browse conversations chronologically by person or topic
• Statistics: Track your conversation patterns and engagement

Configuration

Create a .env file to customize settings:

MODEL_ID=Qwen/Qwen2.5-1.5B-Instruct
USE_GPU=true
MAX_NEW_TOKENS=512
TEMPERATURE=0.7
GOOGLE_CREDENTIALS_PATH=credentials.json

Troubleshooting

Installation Issues

pip install is very slow?

• Use uv instead: uv pip install -r requirements.txt (often 10-100x faster)
• Use pip cache: pip install --cache-dir ~/.cache/pip -r requirements.txt
• Use --prefer-binary flag to avoid building from source
• PyTorch is large (~500MB-2GB). First install takes time, subsequent installs use cache
• For CPU-only systems, use PyTorch CPU build: pip install torch --index-url https://download.pytorch.org/whl/cpu

Network timeouts or slow downloads?

• Use a pip mirror: pip install -i https://pypi.tuna.tsinghua.edu.cn/simple -r requirements.txt (China)
• Increase pip timeout: pip install --default-timeout=1000 -r requirements.txt
• Check your internet connection and firewall settings

Out of disk space?

• PyTorch and dependencies need ~3-5GB free space
• Use pip cache cleanup: pip cache purge (then reinstall with cache)

Permission errors?

• Make sure virtual environment is activated: source .venv/bin/activate
• Don't use sudo with pip in virtual environments

Runtime Issues

Out of memory? Lower MAX_NEW_TOKENS or try USE_GPU=false

Very slow? You're probably on CPU. GPU is much faster.

Script won't run? Make it executable: chmod +x assistant.sh

Google Calendar not working? See GOOGLE_CALENDAR_SETUP.md for complete setup instructions

Documentation

• ASSISTANT.md - Personal Assistant feature documentation
• GOOGLE_CALENDAR_SETUP.md - Google Calendar integration setup guide
• CONTRIBUTING.md - Contribution guidelines

That's It

Your data stays local. No tracking. No cloud calls. Just you and your AI model running on your machine with the power of persistent memory and calendar integration.