SanctuAIry

SanctuAIry is a privacy-first, AI-powered therapeutic assistant. It leverages large language models (LLMs) to simulate therapeutic conversations while prioritizing user confidentiality, privacy, and data security.

Overview

SanctuAIry is designed to run securely in the cloud within confidential virtual machines (VMs), ensuring that sensitive user data—such as therapy sessions and mental health assessments—remains protected and private.

The app aims to provide an empathetic, intelligent, and responsive AI therapist experience by using open-source LLMs, currently powered by Meta’s LLama 3.2 models.

⚠️ While the architecture is planned for deployment in confidential VMs, this feature is not yet implemented. Current development is focused on core functionality.

---

⚙️ Architecture

SanctuAIry is built with a modular architecture, composed of the following key components:

Frontend

A clean and responsive user interface built with React, enabling users to interact with their AI therapist in a seamless way.

LLM Manager

A backend service layer that:

• Routes user input to the appropriate LLM instance
• Manages session context and conversation flow
• Handles model selection, loading, and creation

LLM Core

The brain of the application, where the actual LLM runs:

• Currently utilizes Meta’s LLama 3.2
• Runs in a secured environment to ensure data isolation and confidentiality
• Designed for performance, context awareness, and therapeutic alignment

---

🔒 Privacy & Security

SanctuAIry is committed to privacy by design:

• All processing is planned to occur in confidential virtual machines with hardware-backed security guarantees
• End-to-end encryption for data in transit, in use, and at rest
• Further planed privancy features:
  • GDPR compliance
  • User-defined data retention and deletion policies

---

🚧 Project Status

SanctuAIry is a work in progress. The app is under constant iteration with new features and improvements being added weekly.

Current Focus:

• Improving UI/UX flow and emotional design
• Fine-tuning models for therapy-specific language
• Security hardening and compliance readiness

---

📦 Tech Stack

• Frontend: React and TailwindCSS
• LLM Manager: Node.js and Express
• LLM Core: Ollama, using Meta's LLama 3.2

---

🚀 Running Locally with Docker Compose

To start all services locally using Docker Compose, follow these steps:

1. Prerequisites: Ensure Docker and Docker Compose are installed on your machine.
2. Environment: Copy the example environment file and update the URLs if necessary:

   cp .env.example .env
   # .env:
   # VITE_LLM_MANAGER_URL=http://llm-manager:3000
   # LLM_CORE_URL=http://llm-core:11434
   # NODE_ENV=development
   # PORT=3000
   # OLLAMA_MAX_LOADED_MODELS=2

3. Build & Start: From the project root, run:

   docker-compose up

---

📄 License

License: To Be Decided

This project does not currently have a finalized license. Until then, please reach out before using this code for commercial or derivative purposes.

See the LICENSE file (when available) for future updates.

---

🙏 Acknowledgements

• Meta AI for the open-source LLama models
• The open-source community for tools and inspiration ❤️