PREAA - Providing Reusable Equitable AI Access in Academia

PREAA (Providing Reusable Equitable AI Access in Academia) is a comprehensive open-source platform designed to democratize AI access within academic institutions. Our carefully curated stack of open-source technologies has been researched and selected to provide the most effective, scalable, and equitable AI solutions for educational environments.

🎯 Mission

Academic institutions face significant barriers when implementing AI technologies - from cost constraints to technical complexity. PREAA addresses these challenges by providing a battle-tested, open-source technology stack that enables institutions to deploy enterprise-grade AI capabilities without the typical financial and technical overhead.

🏗️ System Architecture

PREAA integrates several best-in-class open-source technologies to create a comprehensive AI platform:

Core Components

• LibreChat: A ChatGPT-like all-in-one interface providing authentication, conversation history, and multi-model support
• LiteLLM: Unified proxy for LLM requests with rate limiting and standardized OpenAI-compatible interface
• LangFlow: Visual LLM workflow builder supporting multiple AI providers, agentic tools, RAG components, and custom integrations
• LangFuse: Comprehensive LLM analytics platform for token usage tracking, completion costs, and performance metrics

Frontend Components

• Embedded Chat Widget: Lightweight, embeddable chat interface for seamless integration into existing web applications
• Standalone Chat Client: Full-featured React-based chat application with modern UI/UX
• Admin Dashboard: Next.js-based administrative interface for system management and configuration

Backend Services

• Helper Backend: NestJS-based service providing request proxying capabilities and administrative functions
• Custom Integrations:
  • LangFlow custom components for enhanced chat completion interfaces
  • LiteLLM custom providers for workflow integration

Supporting Infrastructure

• PostgreSQL: Primary database for user data, conversation history, and analytics
• MongoDB: Document storage for LibreChat data
• Redis: Caching and session management
• ClickHouse: High-performance analytics database for LangFuse metrics
• MinIO: S3-compatible object storage
• Prometheus & Grafana: Monitoring and visualization stack

🚀 Getting Started

Prerequisites

• Docker and Docker Compose
• Git
• OpenSSL (for key generation)

Quick Start

1. Clone the repository:

   git clone <repository-url>
   cd PREAA

2. Navigate to the deployment directory:

   cd deploy/local

3. Set up environment variables:

   Copy the sample environment files and configure them:

   # Copy all sample environment files
   cp config/.env.clickhouse.sample config/.env.clickhouse
   cp config/.env.minio.sample config/.env.minio
   cp config/.env.redis.sample config/.env.redis
   cp config/.env.psql.sample config/.env.psql
   cp config/.env.langfuse.sample config/.env.langfuse
   cp config/.env.librechat.sample config/.env.librechat
   cp config/.env.litellm.sample config/.env.litellm
   cp config/.env.librechat-metrics.sample config/.env.librechat-metrics
   cp config/.env.langflow.sample config/.env.langflow

4. Configure required secrets:

   Database passwords (generate secure passwords for each):

   • config/.env.clickhouse: Set CLICKHOUSE_PASSWORD
   • config/.env.minio: Set MINIO_ROOT_PASSWORD
   • config/.env.redis: Set REDIS_PASSWORD
   • config/.env.psql: Set POSTGRES_PASSWORD

   LangFuse configuration (config/.env.langfuse):

   # Generate encryption key
   ENCRYPTION_KEY=$(openssl rand -hex 32)
   
   # Set the following variables:
   # SALT=<any secure string>
   # CLICKHOUSE_PASSWORD=<matching value from .env.clickhouse>
   # LANGFUSE_S3_EVENT_UPLOAD_SECRET_ACCESS_KEY=<matching MINIO_ROOT_PASSWORD>
   # LANGFUSE_S3_MEDIA_UPLOAD_SECRET_ACCESS_KEY=<matching MINIO_ROOT_PASSWORD>
   # REDIS_AUTH=<matching REDIS_PASSWORD>
   # DATABASE_URL=postgresql://postgres:<POSTGRES_PASSWORD>@postgres:5432/postgres

   LibreChat configuration (config/.env.librechat):

   # Generate LibreChat secrets using their generator: https://www.librechat.ai/toolkit/creds_generator
   # Set: CREDS_KEY, CREDS_IV, JWT_SECRET, JWT_REFRESH_SECRET
   # Set: LITELLM_API_KEY=<matching LITELLM_MASTER_KEY from .env.litellm>

   LiteLLM configuration (config/.env.litellm):

   # Set: LITELLM_MASTER_KEY=<any secure string>
   # Set: DATABASE_URL=postgresql://postgres:<POSTGRES_PASSWORD>@postgres:5432/postgres

5. Start the platform:

   docker-compose up -d

6. Access the services:

   • LibreChat (Main Interface): http://localhost:3080
   • LangFuse (Analytics): http://localhost:3000
   • LiteLLM (Proxy Management): http://localhost:4000
   • LangFlow (Workflow Builder): http://localhost:7860
   • Grafana (Monitoring): http://localhost:3002
   • Admin Dashboard: Configure separately in packages/admin

Post-Installation Configuration

LangFuse + LiteLLM Integration

1. Create a LangFuse account at http://localhost:3000
2. Set up a new organization and project
3. Generate API keys (public and private)
4. Access LiteLLM at http://localhost:4000/ui
5. Login with username admin and your LITELLM_MASTER_KEY as password
6. Under "Logging & Alerts", add a LangFuse callback
7. Enter your LangFuse API keys and set host to http://langfuse-web:3000
8. Test the integration

🔧 Development

Frontend Development

Each frontend package can be developed independently:

# Embedded Chat Widget
cd packages/embedded-chat
npm install
npm run dev

# Admin Dashboard
cd packages/admin
npm install
npm run dev:https  # Note: Uses HTTPS for functionality

# Chat Client
cd packages/chat-client
npm install
npm run dev

Backend Development

# Helper Backend (NestJS)
cd packages/helper-backend
npm install
npm run start:dev

Custom Components

LangFlow Components:

cd packages/langflow
pip install -r requirements.txt
# Restart LangFlow container to see changes

LiteLLM Custom Providers:

cd packages/litellm
pip install -r requirements.txt
# Restart LiteLLM container to see changes

📊 Monitoring and Analytics

The platform includes comprehensive monitoring capabilities:

• LangFuse: Track LLM usage, costs, and performance metrics
• Prometheus: System metrics collection
• Grafana: Visualization and dashboards
• LibreChat Metrics: Specialized metrics exporter for LibreChat

🔐 Security Considerations

• All services run in isolated Docker containers
• Environment variables store sensitive configuration
• Database connections use authentication
• API keys provide service-to-service authentication
• HTTPS support available for production deployments

🤝 Contributing

We welcome contributions from the academic community! Please see our contribution guidelines and feel free to:

• Report issues or bugs
• Suggest new features
• Submit pull requests
• Share your deployment experiences

📄 License

This project is licensed under the LICENSE file in the repository.

🆘 Support

For support, please:

1. Check the individual component documentation in each package
2. Review the deployment configuration files
3. Open an issue in the repository
4. Consult the upstream documentation for each integrated technology

---

PREAA - Empowering academic institutions with accessible, equitable AI technology.