How does TrustGraph work?

TrustGraph is a context orchestration layer that combines several technologies into a platform designed for accurate, controllable, and explainable knowledge retrieval. Here’s the approach at a high level — later pages go deeper on each aspect.

Multiple retrieval strategies

There is no single retrieval approach that works for every situation. TrustGraph provides three, each suited to different kinds of questions and data:

  • Graph RAG — extracts entities and relationships from documents into a context hypergraph. Queries traverse the graph to find connected information. Best for complex, relationship-rich questions across diverse data
  • Ontology RAG — extends Graph RAG by using a domain ontology to guide extraction (BYOO — Bring Your Own Ontology). You define the entity types and relationships that matter, and TrustGraph extracts accordingly. Best for domain-specific precision
  • Document RAG — the familiar vector similarity approach, enhanced with concept extraction, hybrid retrieval (BM25 + vector fusion), and full explainability. Best for broad semantic search across unstructured content

You choose which strategy fits your use case — or combine them.

HyperFlows: configurable agent workflows

TrustGraph’s architecture is built around HyperFlows — custom agentic workflows where processing capabilities are chained together. Developers can configure specific LLMs and specific hypergraph access permissions for every step of a workflow.

A HyperFlow can route a query from a lightweight local model for classification, to a heavy reasoning model for synthesis, drawing from different hypergraph collections at each step based on governance rules.

HyperFlows are composed of modular services connected through a pub/sub messaging fabric. This means you can:

  • Swap components without changing application code
  • Run multiple retrieval strategies in parallel
  • Add custom processing steps for domain-specific needs
  • Scale individual services independently

Context management

Managing enterprise knowledge at scale requires structure. TrustGraph provides three levels of context organisation:

  • Workspaces — deep, programmatic data isolation for users, agents, and HyperFlows. An HR agent cannot read financial data; multi-tenant data remains strictly compartmentalised
  • Collections — distinct knowledge bases within a workspace that can be partitioned, managed, and dynamically combined at query time
  • Context Cores — modular, portable, reusable units of context. Package domain-specific knowledge into a Context Core and plug it into any agent or workflow. Context Cores load in a fraction of the time taken to create the original knowledge

Open LLM inference stack

TrustGraph works with over 40 LLM providers — including OpenAI, Anthropic, Google VertexAI, AWS Bedrock, Mistral, Ollama, and vLLM. It’s not locked into any single provider or deployment model.

You can use cloud-hosted models, self-hosted open-source models on any hardware (Nvidia, AMD, or Intel accelerators), or a mix. The same HyperFlow works regardless of which LLM is behind it — keeping your data and compute entirely within your sovereignty.

Explainability built in

TrustGraph captures all event metadata in the hypergraph, providing real-time traceability for every decision an agent makes. Which concepts were extracted from the question, which hypergraph nodes were visited, which edges were selected and why, and which source documents contributed to the answer.

This isn’t a bolt-on feature — it’s woven into every retrieval pipeline. If an agent takes an action, you can trace the exact path through the hypergraph that led to that outcome.

Open and extensible

TrustGraph is fully open source with a modular architecture. All services communicate through open APIs, so you can extend the platform with custom processors, tools, and integrations without modifying core code.

Next

Retrieval methodologies — a closer look at how Graph RAG, Document RAG, and Ontology RAG work and when to use each.