TrustGraph
Back to News
Release 2.1

TrustGraph Launches Version 2: End-to-End Explainability and Context Graph-Native AI Infrastructure for Production

March 28, 2026
4 min read

San Francisco, CA — TrustGraph, the open-source context development platform for AI applications, today announces the official release of TrustGraph 2, bringing enterprise-grade explainability, provenance tracking, and the leading context graph architecture to developers building context-aware AI systems.

TrustGraph 2 represents a foundational step forward for teams who need more than retrieval — they need to understand why their AI says what it says.

Full-Stack Explainability, From Document to Answer

The headline capability of TrustGraph 2 is end-to-end explainability across the entire knowledge pipeline. Every step of the process — from raw document ingestion through chunk extraction, graph construction, and multi-stage query reasoning — is now traceable and auditable.

At extract time, TrustGraph 2 emits PROV-O triples that trace the lineage of every piece of knowledge from its source document, through pages and chunks, down to individual graph edges. At query time, GraphRAG, DocumentRAG, and Agent queries record full reasoning traces — capturing question formation, grounding, exploration, focus, and synthesis stages — into a dedicated named graph. Developers can inspect these traces directly from the CLI using new tools like tg-list-explain-traces and tg-show-explain-trace, or view them inline in the TrustGraph Workbench's new Explainability Panel.

"AI systems that can't explain themselves aren't production-ready," said TrustGraph Co-Founder Mark Adams. "TrustGraph 2 makes explainability a first-class citizen — not an afterthought."

A Redesigned Knowledge Graph Architecture Becomes Context Graphs

TrustGraph 2 introduces a new entity-centric graph schema alongside a redesigned core wire format built on typed RDF Terms. The new Term format replaces the legacy Value schema with full support for IRIs, typed literals, RDF-star quoted triples, and blank nodes — aligning TrustGraph with semantic web standards and enabling richer, more expressive knowledge representations.

Context is now organized across named graphs: the default graph for facts, a source provenance graph, and a retrieval explainability graph. This separation makes it possible to query, audit, and export knowledge with full context about where it came from and how it was used.

Pluggable Tools and Smarter Agent Frameworks

TrustGraph 2 ships with a new Tool Services architecture, enabling developers to build and register custom tool implementations that agent frameworks can discover and invoke dynamically at runtime. Tools can be deployed independently and composed into agent workflows without hardcoded integrations — a significant step toward modular, production-grade agentic systems.

Batch Embeddings and Streaming Retrieval at Scale

The embeddings service has been redesigned for batch processing, with all providers — including FastEmbed and Ollama — updated to accept lists of texts and return similarity scores. A new suite of CLI tools gives developers direct access to graph, document, and row-level embeddings queries.

For large knowledge bases, TrustGraph 2 introduces streaming triple queries with configurable batch sizes, reducing time-to-first-result and memory overhead. Large document ingestion now supports multipart S3 upload with streaming retrieval, making it practical to load and process enterprise-scale document libraries.

TrustGraph Workbench 1.7.2

The TrustGraph Workbench — the visual interface for managing and querying knowledge graphs — ships version 1.7.2 alongside TrustGraph 2.1. The update adds the Explainability Panel to the Assistant chat, letting users inspect reasoning traces and provenance without leaving the browser. Row embedding support and full compatibility with the TrustGraph 2 API round out the workbench release.

Open Source and Self-Hostable

TrustGraph remains fully open source and designed to run anywhere — locally or in the cloud — with no vendor lock-in. Like Supabase for application databases, TrustGraph provides the full developer toolkit for context engineering out of the box: multi-model storage, semantic retrieval pipelines, portable context cores, and now a complete provenance and explainability layer.

TrustGraph 2 is available today at https://trustgraph.ai and on GitHub.

For more information:

View all news articles