TrustGraph Containers

TrustGraph uses a modular container architecture where different containers provide specialized capabilities. This approach allows for flexible deployment where only needed capabilities are included, reducing resource usage and attack surface while maintaining full functionality when all containers are deployed together.

Version strategy

The TrustGraph release process creates containers and packages with the same version number. For best results, match the container and package version numbers.

Container Overview

Container Purpose Key Features Use Case
trustgraph-base Foundation container with basic building blocks, client APIs, and base classes • Core Python runtime (3.12)
• Basic HTTP client capabilities (aiohttp)
• Pulsar messaging system integration
• Foundational libraries for other containers		 Required as a base layer for other containers. Contains minimal dependencies focused on core messaging and HTTP capabilities.
trustgraph-flow Main processing container containing the bulk of TrustGraph’s capabilities • Multi-provider AI integration (OpenAI, Anthropic, Cohere, Mistral, Google Generative AI, Ollama)
• LangChain ecosystem with complete text processing
• Database support (Milvus, Neo4j, FalkorDB, Cassandra)
• RDF/Semantic web capabilities
• Document processing and analysis		 Core container for most TrustGraph workflows. Deploy when you need full AI processing capabilities, document handling, or database integration.
trustgraph-mcp Model Context Protocol (MCP) server functionality • MCP server implementation
• WebSocket-based communication
• Lightweight protocol handling		 Deploy when you need MCP server capabilities for model context management and protocol-based communication.
trustgraph-hf Hugging Face model processing with local ML inference • PyTorch support (CPU-optimized)
• Hugging Face integration (Transformers, sentence transformers, embeddings)
• Local ML inference without external API calls
• Pre-loaded models (all-MiniLM-L6-v2)		 Deploy when you need local ML model inference, text embeddings, or want to avoid external API dependencies for certain AI tasks.
trustgraph-ocr Optical Character Recognition and document processing • Tesseract OCR for text extraction from images
• PDF processing with Poppler utilities
• Complete document processing pipeline		 Deploy when you need to process scanned documents, extract text from images, or handle PDF document analysis.
trustgraph-bedrock AWS Bedrock AI services integration • AWS Bedrock model access
• Cloud-based AI inference
• Lightweight AWS-specific integration		 Deploy when using AWS Bedrock as your AI provider. Provides dedicated integration without the overhead of other AI providers.
trustgraph-vertexai Google Vertex AI integration • Google Cloud Vertex AI model access
• Cloud-based AI inference
• Google AI Platform SDK integration		 Deploy when using Google Vertex AI as your AI provider. Provides dedicated integration for Google’s AI/ML platform.

Architecture Principles

Modular Design

Each container is purpose-built for specific AI providers or capabilities. This allows you to:

  • Mix and match containers based on deployment needs
  • Reduce resource usage by only including necessary dependencies
  • Minimize attack surface by avoiding unused components
  • Scale individual components independently

Common Foundation

All containers share common patterns:

  • Base OS: Fedora 42 for security and stability
  • Python Runtime: Python 3.12 for modern language features
  • Messaging: Pulsar messaging system for distributed communication
  • Build Strategy: Multi-stage builds for optimized container sizes

Deployment Flexibility

Minimal Deployment:

  • trustgraph-base as a base for extension
  • trustgraph-flow for the most common AI capabilities
  • trustgraph-mcp for MCP protocol suppport

Document Processing:

  • Add trustgraph-ocr for document OCR with Tesseract

Local ML Processing:

  • Add trustgraph-hf for local model inference without external APIs

Cloud AI Integration:

  • Add trustgraph-bedrock for AWS Bedrock
  • Add trustgraph-vertexai for Google Vertex AI (Google AIStudio is supported in trustgraph-flow.

Container Dependencies 

trustgraph-base (foundation)
├── trustgraph-flow (most of the capability is here)
├── trustgraph-hf (HuggingFace, local ML, transformers model)
├── trustgraph-ocr (tesseract)
├── trustgraph-bedrock (AWS Bedrock)
└── trustgraph-vertexai (Google AI with VertexAI libraries)
trustgraph-mcp (MCP protocol server)

Most containers depend on trustgraph-base for core functionality, while specialized containers can be deployed independently based on your specific requirements.