TrustGraph
Back to News
Release 2.2

TrustGraph Releases Version 2.2 with Multi-Pattern Agent Orchestration, RabbitMQ Support, and Native SPARQL Query Service

April 8, 2026
4 min read

San Francisco, CA — TrustGraph, the open-source context operating system for AI agents, today announce the release of version 2.2, delivering a powerful multi-pattern agent orchestrator, a pluggable pub/sub messaging layer, and a backend-agnostic SPARQL 1.1 query service. The release marks a significant step forward in TrustGraph's mission to provide deterministic, explainable AI infrastructure for enterprise and open-source deployments alike.

Multi-Pattern Agent Orchestrator

At the heart of v2.2 is a new Agent Orchestrator that brings LLM-based meta-routing to TrustGraph. Rather than locking users into a single execution strategy, the orchestrator intelligently selects the most appropriate pattern per request from three supported modes:

  • Plan-then-Execute — the LLM generates a structured plan of steps, executes each sequentially, and synthesizes results into a final response
  • Supervisor — decomposes a question into sub-agent goals, fans out to parallel sub-agents, and aggregates findings into a unified synthesis
  • ReAct — the existing iterative reasoning pattern, unchanged and fully supported

Full explainability provenance is maintained across all three patterns using new RDF types (Decomposition, Finding, Plan, StepResult, Synthesis) and predicates (tg:subagentGoal, tg:planStep). The orchestrator also ships with over 96 tests covering aggregation, provenance, routing, explainability parsing, DAG structure, and callback message IDs. Pattern selection is available directly from the CLI via tg-invoke-agent -p supervisor|plan-then-execute|react.

RabbitMQ Pub/Sub Backend

Version 2.2 introduces a pub/sub abstraction layer that decouples TrustGraph from any single messaging fabric. RabbitMQ is now available as an alternative to Apache Pulsar, selected for its significantly lower resource requirements — making TrustGraph more accessible for teams who don't need Pulsar's scale. The backend is selectable via the PUBSUB_BACKEND=rabbitmq environment variable, with topic exchange architecture, subscriber resilience through automatic consumer recreation, and a thread-safe consumer model. Support for additional fabrics, including Kafka, is planned for a subsequent release.

SPARQL 1.1 Query Service

TrustGraph v2.2 introduces a backend-agnostic SPARQL 1.1 query service that parses queries using rdflib and decomposes them into triple pattern lookups via the existing pub/sub interface. The service supports a comprehensive feature set including BGP, JOIN, OPTIONAL, UNION, FILTER, BIND, VALUES, GROUP BY, ORDER BY, LIMIT/OFFSET, DISTINCT, and aggregates. Batching and streaming support handles large result sets gracefully. The service is accessible via the API gateway, Python SDK (FlowInstance.sparql_query), and CLI (tg-invoke-sparql-query).

Universal Document Decoder

A new Universal Document Decoder consolidates multi-format document processing into a single service powered by the unstructured library. Supported formats include DOCX, XLSX, PPTX, HTML, Markdown, CSV, RTF, ODT, EPUB, and more. Tables are preserved as HTML markup, images are routed to the librarian, and configurable section grouping strategies give teams fine-grained control over how documents are chunked and stored.

Additional Improvements

  • Inline Explainability Triples — Provenance triples are now included directly in explain messages from GraphRAG, DocumentRAG, and Agent services, eliminating follow-up knowledge graph queries for explainability details
  • Persistent WebSocket Connections — A single persistent connection with request multiplexing replaces per-request connections, eliminating repeated TCP+WS handshakes and significantly reducing overhead
  • Config Push Notify Pattern — Lightweight version-based config signals replace stateful config broadcasts for improved efficiency
  • Auto-pull Ollama Models — The Ollama provider now automatically pulls missing models on first use
  • MCP Gateway Auth — GATEWAY_SECRET environment variable support for secure MCP server to API gateway authentication
  • Prompt Queue Monitoring — New tg-monitor-prompts CLI tool for subscribing to prompt request/response queues with correlation and timing summaries

Breaking Changes

Teams upgrading to v2.2 should note the following breaking changes:

Pub/sub queue naming has changed from a topic-based format to CLASS:TOPICSPACE:TOPIC; translator methods have been renamed from to_pulsar/from_pulsar to encode/decode

Agent schema has been updated with orchestration fields; legacy response fields (answer, thought, observation) have been removed

Config push schema now contains a types list instead of the full config dict; the state queue class has been replaced by the flow class

Full migration guidance is available in the TrustGraph documentation and GitHub release notes.

Availability

TrustGraph v2.2 is available now on GitHub. The project is open source under the Apache 2.0 license. Documentation, quickstart guides, and deployment resources are available at trustgraph.ai.

For more information:

View all news articles