TrustGraph 2.8 Launches with Async Architecture, Hybrid Retrieval and Structured AI Output

TrustGraph has released version 2.8 of its enterprise knowledge and AI platform, introducing a fully asynchronous messaging infrastructure that replaces the previous thread-per-consumer model, reducing resource overhead in multi-workspace deployments. The update adds hybrid document retrieval combining vector similarity search with BM25 keyword matching, merged via Reciprocal Rank Fusion, with SQLite FTS5 as the initial keyword-index backend. Structured output support now carries JSON schemas through to provider-native APIs across more than a dozen LLM backends, enabling typed, machine-readable responses without post-generation parsing. The release also introduces a pluggable image-to-text service, a Docling-powered document decoder supporting formats including PDF, DOCX, and PPTX, and expanded metrics coverage using a unified tg_ prefix across the platform. Audit logging has been strengthened with structured gateway and IAM events, including request and client-IP correlation for compliance and observability purposes.