Atarax-AI

A Local, Privacy-Preserving AI Assistant Powered by llama.cpp

Fully offline. Multi-modal. Secure. Yours.

Atarax-AI is a full-featured AI assistant that runs entirely offline using llama.cpp, optimized for low-latency, high-accuracy inference on consumer-grade hardware. The assistant supports multi-modal inputs (text, voice, images, and videos), performs real-time reasoning, and integrates with local system APIs (all with zero cloud dependency).

Architecture Overview

Note: We are currently refactoring the backend before integrating the front-end.

Project Vision

Create a production-ready AI assistant that:

• 100% Offline Operation - No internet connection required after setup
• Privacy-First Design - All data processing happens locally
• Consumer Hardware Optimized - Runs efficiently on laptops and desktops
• Multi-Modal Interaction - Text, voice, and document processing capabilities
• Seamless Integration - Works with your existing workflow and applications

Key Features

Core Capabilities

• Intelligent Text Assistant - Context-aware responses with advanced reasoning
• Voice Interaction - Whisper.cpp integration for natural speech processing
• Document Processing - Local file analysis and knowledge base creation
• Persistent Memory - Semantic search with long-term context retention
• System Integration - Calendar, file management, and task automation

Technical Excellence

• Adaptive Context Management - Sliding window techniques for optimal performance
• Modular Architecture - Flexible prompt engineering framework
• Real-Time Monitoring - Performance optimization and comprehensive logging
• Cross-Platform Support - Linux, macOS, and Windows compatibility

Security & Privacy

• Local Key Management - Password-derived keys, never stored on disk
• Secure Data Deletion - Cryptographic erasure of sensitive information
• Zero Telemetry - No data collection or external transmissions

Quick Start

We will provide soon packages for Linux/Windows/Mac. For now, you can build the docker locally. Optionnaly you can pull it from docker hub (https://hub.docker.com/repositories/ataraxai).

Prerequisites

• Docker and Docker Compose
• NVIDIA GPU (optional, for GPU acceleration)
• 8GB+ RAM recommended

Option 1: Docker (Recommended)

Pull from Docker Hub

# For the CPU version
docker pull ataraxai/ataraxai:latest
docker run -it --rm -p 8000:8000 ataraxai/ataraxai:latest

CPU Version locally

# Build the image
docker build -t ataraxai:cpu -f docker/Dockerfile.cpu .

# Run the container
docker run -it --rm -p 8000:8000 ataraxai:cpu

GPU Version locally

# Build the image
docker build -t ataraxai:gpu -f docker/Dockerfile.gpu .

# Run with GPU support
docker run --gpus all -it --rm -p 8000:8000 ataraxai:gpu

Option 2: Local Installation

# Run the installation script
./install.sh

# Available options:
# --use-uv          This will create a .venv to use uv env (You need to have uv install in your OS) 
# --use-conda       Use conda environment
# --clean           Clean previous builds
# --clean-ccache    Clean C++ compilation cache
# --use-cuda        Build with CUDA support
# --only-cpp        Build only C++ components
# --cuda-arch       Specify CUDA architecture

# Start the API server
uvicorn api:app --reload --host 0.0.0.0 --port 8000

Monitoring & Observability

Launch the monitoring stack to track performance and resource usage:

Full Monitoring (CPU + GPU)

docker compose -f docker-compose.monitoring.base.yml -f docker-compose.monitoring.gpu.yml up -d

CPU-Only Monitoring

docker compose -f docker/docker-compose.monitoring.base.yml up -d

Access Monitoring Services

• FastAPI Backend: http://localhost:8000
• Prometheus Metrics: http://localhost:9090
• Node Exporter: http://localhost:9100
• Grafana Dashboard: http://localhost:3000

Documentation

API Reference

Access the interactive API documentation at http://localhost:8000/docs once the server is running.

Privacy Commitment

Atarax-AI is designed with privacy as a fundamental principle:

• No Cloud Dependency - Everything runs on your hardware
• No External API Calls - Zero network requests to third-party services
• Local Data Storage - All data encrypted and stored locally
• Open Source Transparency - Full source code inspection and auditing
• User Control - Complete ownership of your data and interactions

Contributing

We welcome contributions! Please see our Contributing Guidelines for details on:

• Code style and standards
• Testing requirements
• Pull request process
• Issue reporting

Roadmap

• Desktop application with Tauri frontend
• Optimization
• Mobile application support
• Plugin system for custom integrations
• Advanced document understanding
• Multi-language support
• Voice synthesis capabilities

License

This project is licensed under the GNU GPLv3 License. You are free to use, modify, and distribute this software under the terms of the GPL.

Acknowledgments

Built with these amazing open-source projects:

• llama.cpp - High-performance LLM inference
• Whisper.cpp - Speech recognition
• FastAPI - Modern Python web framework
• Tauri - Cross-platform desktop applications

---

Made with ❤️ for privacy-conscious users who value local AI