trustgraph-ai/trustgraph

The Governance Platform for Agents, Knowledge, and Context

TrustGraph – Ontology‑Native Semantic Intelligence Platform

What it is

  • An open‑source platform that turns raw documents (PDF, DOCX, CSV, images, etc.) into a structured, provenance‑rich knowledge graph built on W3C standards (RDF 1.2, OWL, PROV‑O).
  • The resulting graph can be queried with natural language, SPARQL, or GraphRAG and is designed to feed LLM‑based agents with unambiguous, traceable facts.

Why it matters

  • Conventional RAG pipelines rely on vector similarity over isolated text chunks, which often leads to hallucinations because the model cannot distinguish meaning (e.g., “Who” as a person vs. a question).
  • TrustGraph injects a formal ontology into the retrieval pipeline, so every entity and relationship is typed, constrained, and linked to its source document.
  • This gives AI agents a shared, machine‑readable context, dramatically improving consistency and explainability.

Core components

Component Role
Semantic Intelligence Layer RDF 1.2 named‑graph store with automated NL‑to‑graph retrieval, semantic filtering and reranking.
Ontology & Schema Workbench Visual editor for OWL ontologies, import/export (Turtle, OWL/XML), validation and dependency checking.
Agent Runtime & API Gateway Native runtime that records every agent action, links decisions back to provenance triples, and exposes a REST/GraphQL‑style API.
Knowledge Cores / Workspaces Isolated collections that can be versioned, accessed with fine‑grained ACLs, and used as reusable “knowledge packages.”
Ingestion Pipelines Built‑in converters for PDFs, DOCX, XLSX, PPTX, HTML, Markdown, CSV and OCR‑extracted images, mapping extracted entities to ontology classes.
LLM Stack Connectors for all major provider APIs (OpenAI, Anthropic, Cohere, Gemini, Mistral) and self‑hosted models via vLLM, TGI, Ollama, LM Studio, Llamafile.
Infrastructure Services Managed storage back‑ends: Cassandra (multi‑model), Qdrant (vector embeddings), Garage (S3‑compatible object store), Pulsar or RabbitMQ for pub/sub.

Key features

  • Standard‑compliant: All data lives in RDF/OWL, exportable to any W3C‑compatible system.
  • Provenance‑first: Extraction lineage and agent query traces are stored as PROV‑O triples, enabling full auditability.
  • Natural‑language retrieval: Users can ask questions in plain English; the system translates them to SPARQL/GraphRAG against the ontology‑enhanced graph.
  • Modular deployment: Distributed as Docker containers (or Kubernetes resources) with a single‑command config wizard (npx @trustgraph/config).
  • Developer libraries: TypeScript client (@trustgraph/client) and React state/provider packages for rapid UI integration.
  • Zero‑key bootstrap: Only LLM/OCR provider keys are optional; the platform itself runs without additional secrets.

Typical use cases

  • Enterprise knowledge bases where regulatory compliance demands traceable source attribution.
  • RAG for complex domains (legal, medical, finance) where precise entity types and relationships matter.
  • Semantic search assistants that need to answer queries with explicit reasoning paths.
  • AI‑agent orchestration where multiple agents share a common ontology to avoid miscommunication.

Tech stack

  • Data model: RDF 1.2 quads, OWL 2, PROV‑O.
  • Storage: Apache Cassandra (key‑value), Qdrant (vector), Garage (S3‑compatible), Apache Pulsar / RabbitMQ (messaging).
  • LLM inference: vLLM, Text Generation Inference, Ollama, LM Studio, Llamafile.
  • APIs: RESTful API gateway, optional GraphQL, SPARQL endpoint.
  • UI: Docker‑served web UI (port 8888) with React components for agent console, GraphRAG view, 3‑D context explorer, ontology editor, etc.
  • Packaging: Docker Compose, Helm/Kubernetes manifests, CLI config tool.

Getting started

  1. Install Docker (or Podman/Minikube).
  2. Run the one‑liner config wizard: npx @trustgraph/config.
  3. The wizard produces a deploy.zip containing either a docker‑compose.yaml or a resources.yaml for Kubernetes.
  4. Deploy the stack, then open the UI at http://localhost:8888.
  5. Use the Document Ingestion UI or CLI to upload source files; select or create an OWL ontology; the system will extract typed triples automatically.
  6. Query via the Agent Console or the REST API; connect your own LLM‑based agent through the API gateway.

Community & support

  • Discord community for questions, bug reports and feature requests.
  • Full documentation site (getting‑started guides, API reference, deployment manuals).
  • Open‑source under Apache 2.0 (2024‑2026 TrustGraph).

Bottom line TrustGraph provides the missing “semantic intelligence” layer that lets LLM‑driven applications work with a rigorously defined knowledge graph instead of raw text chunks, delivering more reliable, explainable, and provenance‑aware AI results.

Related

  • Project
    infiniflow/ragflow

    RAGFlow is an open‑source Retrieval‑Augmented Generation platform that adds agentic reasoning to LLMs. It ingests many document types, creates searchable embeddings, and supports multi‑step retrieval with configurable “thinking” depth. Features include template‑based chunking, knowledge compilation (wikis, graphs, mind‑maps), grounded citations, and plug‑in LLM/embedding models. Deploy locally via Docker or from source; a hosted cloud version is also offered. Licensed under Apache‑2.0 and backed by an active community.

  • Project
    monarch-initiative/ontogpt

    A Python package for extracting structured, ontology-grounded information from text using LLMs, preventing hallucinations by validating identifiers against real-world ontologies.

  • Project
    iMoonLab/Hyper-RAG

    A hypergraph-driven Retrieval-Augmented Generation (RAG) framework that reduces LLM hallucinations by capturing complex, beyond-pairwise relationships in domain-specific knowledge.

  • Project
    databrickslabs/ontobricks

    OntoBricks is a Databricks-native app that turns Unity Catalog tables into a living, queryable knowledge graph using LLM-driven ontology generation and mapping. It provides governance, multiple graph backends, and exposes the graph to AI agents via MCP.

  • Project
    FalkorDB/GraphRAG-SDK

    A modular framework for building GraphRAG applications on FalkorDB that improves factual accuracy and reduces LLM hallucinations by using knowledge graphs for traceable, multi-hop evidence retrieval.