Streaming Responses for Ontology-Driven AI in TrustGraph Release 1.6

Streaming Responses for Ontology-Driven AI in TrustGraph Release 1.6

TrustGraph release 1.6

SAN FRANCISCO, CA — December 4, 2025 — TrustGraph, the open-source agentic AI platform built for enterprises, today announces the release of version 1.6, introducing comprehensive streaming support that transforms real-time interactions with large language models, knowledge retrieval systems, and multi-step AI agents.

Streaming as a First-Class Capability
Version 1.6 marks a significant milestone in TrustGraph's evolution: streaming is now a native, production-ready feature across the entire platform. This release addresses a critical gap in enterprise AI infrastructure—the ability to deliver real-time, token-by-token responses that reduce latency, improve user experience, and enable truly interactive agentic workflows.

Full Streaming Support Across All LLM Providers
The release introduces comprehensive streaming infrastructure integrated with every major LLM provider, including Azure, Claude, Cohere, Google AI Studio, Mistral, OpenAI, Vertex AI, and private model services like Ollama, LM Studio, Llamafile, and TensorFlow Text Generation Inference (TGI).

Developers can now configure streaming at the schema level and invoke it seamlessly across TrustGraph's full API surface—REST, WebSocket, Python client, and CLI tools. Critically, all streaming implementations are fully backward compatible with existing non-streaming clients, ensuring a smooth transition for production deployments.

Extended Streaming to GraphRAG and Document Retrieval
TrustGraph 1.6 extends streaming beyond raw LLM completions to knowledge-driven retrieval systems. GraphRAG, driven by ontologies, and DocumentRAG services now stream token-by-token responses, delivering insights from knowledge graphs and document stores in real-time. Queries that previously waited for complete batch processing now return results incrementally, enabling enterprises to surface knowledge faster and improve responsiveness in mission-critical applications.

The implementation maintains consistency across all TrustGraph services—developers experience a unified streaming UX whether querying an LLM directly, retrieving from a knowledge graph, or invoking a multi-step agent workflow.

Real-Time Agent Interactions and Streaming Thought Process
Enhanced streaming support for agentic workflows represents a watershed moment for transparent, observable AI reasoning.

  • Observe reasoning in flight — Watch agents think through problems step by step

  • Deliver incremental results — Surface intermediate conclusions as agents work through multi-step workflows

  • Build transparent interfaces — Create user-facing applications that show AI reasoning progressively, building trust through explainability

  • Reduce apparent latency — Deliver immediate feedback during long-running agent operations

The release includes a robust streaming parser for agent responses, with comprehensive error handling and automatic recovery mechanisms designed for production reliability.

Gateway API and WebSocket Support
The API Gateway now natively supports WebSocket streaming, enabling real-time client applications to receive token-by-token responses without polling or long-polling inefficiencies. This architecture scales effortlessly across multiple concurrent connections, crucial for enterprise deployments serving thousands of users.

Enhanced Integration Testing
Comprehensive integration tests validate streaming functionality across LLM providers, RAG systems, and agent workflows, ensuring consistency and reliability in production environments.

Critical Bug Fixes and Compatibility Improvements
AWS Bedrock Model Invocation — Resolved compatibility issues with newer Bedrock model invocation APIs, including full streaming support for the latest Bedrock models.

Minio Library Compatibility — Fixed incompatible library changes in the Minio client for blob storage operations, ensuring stable object storage interactions in deployments using S3-compatible backends.

Streaming Agent Race Conditions — Eliminated race conditions and message ordering issues in streaming agent responses, guaranteeing deterministic, auditable agent behavior—a critical requirement for enterprise compliance.

Developer Experience Enhancements
New CLI improvements expand TrustGraph's developer tooling. The tg-dump-queues utility now provides diagnostic insights into streaming queue states, enabling engineers to troubleshoot and optimize queue configurations in real-time.

Deployment Flexibility and Model Support
Version 1.6 includes updated templates for the latest Bedrock and Claude models, ensuring seamless integration with state-of-the-art LLMs. The platform continues to support deployment across cloud providers (AWS, Azure, Google Cloud, OVHcloud, Scaleway), private data centers, and bare-metal infrastructure.

Building Production-Grade Agentic AI
TrustGraph's philosophy remains unchanged: build infrastructure that enables enterprises to move from AI demos to durable, mission-critical systems. Streaming in 1.6 is not a bolt-on feature—it's architected into the core platform, designed for production use from day one.

The combination of Apache Pulsar's event-driven backbone, knowledge graph semantics, and now comprehensive streaming support creates an AI infrastructure stack uniquely suited to enterprises that refuse to compromise on data sovereignty, auditability, or reliability.

Availability
Version 1.6 is available immediately via GitHub at https://github.com/trustgraph-ai/trustgraph and through the Configuration Builder at https://config-ui.demo.trustgraph.ai/.

The release is fully open source under the Apache 2.0 license, enabling enterprises to deploy, modify, and operate the platform on their own infrastructure with complete transparency and control.

For more information: