TrustGraph 2.6 Ditches Slow LLM Scoring for Faster Open-Source Knowledge Graphs
TrustGraph 2.6 ships FlashRank cross-encoder reranking, portable workspace bundles, and fine-grained enterprise IAM for graph-native agent memory
ProTakeaways
- TrustGraph 2.6 is out, replacing its LLM-based edge scoring with a FlashRank cross-encoder reranker for faster, more accurate GraphRAG retrieval.
- New portable
.tgxworkspace bundles let teams export and import full graph deployments (configs, triples, documents) via CLI. - Multi-provider LLM support via a single
--variantflag covers DeepSeek, Qwen, Mistral, LLaMA, and more through one OpenAI-compatible processor. - Fine-grained IAM splits broad permissions into per-operation capabilities (e.g.,
sparql:read,triples:read) for multi-tenant enterprise deployments. - A major UI overhaul adds domain-specific graph explorers including pathway finders, risk viewers, and a game theory visualizer with Nash Equilibrium detection.
- Free, Apache 2.0 open source, self-hosted on GitHub with support for Neo4j, Cassandra, Qdrant, and 40+ LLM providers.
TrustGraph is an open-source platform that takes a fundamentally different approach to grounding AI agents: instead of handing an LLM a bag of text chunks, it builds a structured knowledge graph from your data and lets agents query it with the precision of a database. Version 2.6 is the biggest release yet, overhauling how the system ranks and retrieves information, making workspaces portable, and adding enterprise-grade access control.
The core problem it solves
Standard RAG (Retrieval-Augmented Generation) works by converting documents into vector embeddings and fetching the chunks most similar to a query. The problem is that similarity is not the same as relevance, and flat chunks lose the relationships between entities. Agents need to retrieve relevant past work, preserve relationships among entities, and distinguish durable knowledge from temporary observations -- and knowledge graphs are well suited to this because they store information as entities and relationships rather than as isolated text chunks.
TrustGraph is an open-source context development platform that combines data streaming, knowledge graph construction, vector search, and LLM orchestration into a single containerized platform designed to reduce hallucinations and improve the accuracy of AI systems. The key idea is that every fact lives in a graph as a typed triple -- Subject, Predicate, Object -- governed by an OWL ontology (a formal schema that defines what each term means). When TrustGraph extracts knowledge using an OWL ontology, every entity has a type from the ontology's class hierarchy and every relationship has a property from the ontology's property definitions -- so an LLM reading the graph can trust that "diagnosis" refers to a clinical diagnosis, not a software bug diagnosis, because the ontology enforces it.