TrustGraph

TrustGraph is a context orchestration layer that turns raw enterprise data into structured context for agents. It ingests unstructured sources such as documents, wikis, APIs, and databases, extracts entities and relationships with language models, and stores them as a hypergraph rather than relying on vector similarity alone.

The hypergraph is built on RDF 1.2 with named graphs serialised as N-Quads, so a whole statement can be referenced as a node and a multi-entity event can be grouped into one addressable unit. Where a standard knowledge graph links a document to an author, this structure connects the document, author, approving manager, compliance policy, and time and location metadata as a single relational event. Ontologies supplied in OWL govern semantic compliance for ingested data, and agents traverse the result with SPARQL or GraphRAG.

Context is managed in three units. Workspaces isolate data programmatically for users, agents, and workflows. Collections are distinct knowledge bases that can be partitioned, queried separately, or combined at query time. Context Cores are portable, reusable packages of domain knowledge that plug into an agent or workflow at runtime. Hyperflows chain processing steps, so one query can route from a lightweight local model for classification to a heavier reasoning model, drawing on different collections at each step.

Configuration produces a deployment bundle for Docker, Podman, or Kubernetes, and the same configuration can be done in a browser terminal. The stack bundles Cassandra for multi-model storage, Qdrant for vector embeddings, Garage for S3-compatible object storage, and Pulsar or RabbitMQ for messaging, plus an inference stack for open models on vLLM, TGI, Ollama, LM Studio, and llamafiles across Nvidia, AMD, and Intel accelerators. It runs self-hosted and air-gapped, inside a customer's own cloud, or as a managed service.

Event metadata is captured in the hypergraph, so the path that led to an agent's action can be traced after the fact. A web console on port 8888 covers the major features, and three TypeScript packages exist for embedding TrustGraph services in other interfaces.

Features

• Hypergraph context: RDF 1.2 and named graphs as N-Quads, capturing n-ary relationships between entities and events
• Bring your own ontology: OWL ontologies enforce semantic compliance on ingested data, with automated ontology-compliant retrieval
• Automated graph construction: language models extract entities and relationships from unstructured sources into the graph
• Context Cores: modular, portable context packages loaded and removed at runtime
• Workspaces and collections: per-user and per-agent isolation, and separately managed knowledge bases combined on demand
• Hyperflows: chained agentic workflows with per-step model and graph configuration under governance rules
• Provenance: event metadata in the hypergraph gives a traceable path behind each agent decision
• Open inference stack: vLLM, TGI, Ollama, LM Studio, and llamafiles on Nvidia, AMD, or Intel hardware
• Bundled storage: Cassandra, Qdrant, Garage object storage, and Pulsar or RabbitMQ messaging included
• Web console: agent console, GraphRAG view with an explainability DAG, 3D context explorer, and ontology and schema workbenches
• TypeScript libraries: client, React state, and React provider packages for building interfaces
• Deployment options: self-hosted or air-gapped, into your own cloud, or as a managed service