GraphRAG 1.2.0 Release Notes

05/06/2026

This document outlines the new features, enhancements, bug fixes, and updated compatibility information for the GraphRAG application, minor version 1.2.0. This release strengthens the end-user experience and trust layer (guardrails and clearer messaging), improves semantic explainability (taxonomy concept enrichment), and hardens graph-tool integrations for more reliable enterprise demos and deployments.

Important

All references to PoolParty have been updated to Graph Modeling across workflows, scripts, and documentation.

Expanded SKOS concept enrichment and UI display for better explainability.
Significantly improved guardrails behavior for off-topic/insufficient-context questions
More robust GraphDB MCP behavior (timeouts and prompt refinements).
Vector search is now available as an MCP tool so the MCP client can route calls intelligently (reduces unnecessary parallel calls).
Configuration hardening: Embeddings Model ID is now centralized and required; vector search validates minimum input length.
Context-aware follow-up questions.
UI updates: Graphwise look and feel improvements and rebranding.

Richer Concept Details from Graph Modeling and UI Extension

Responses now include additional SKOS attributes, allowing concept enrichment to provide a more comprehensive semantic view from Graph Modeling in the UI, beyond just labels and definitions. The UI has been updated to display these additional SKOS properties for each highlighted concept. This improvement adds deeper semantic context beyond prefLabel/altLabel/definition as well as transparency and explainability for concept grounding.

Improved Output Guardrails and Context Awareness

The output guardrails and final answer prompting have been significantly improved for handling off-topic questions. Guardrails now provide descriptive explanations when triggered, appropriately suppress follow-up questions, and ensure that no source documents are returned when context is insufficient. Additionally, guardrails no longer misinterpret dataset content as false data.

These improvements increase user trust by delivering clear, consistent messaging when questions fall outside the dataset's scope. Off-topic and out-of-context queries now produce reliable, well-explained responses instead of inconsistent or misleading answers.

Vector Search as an MCP Tool

The n8n workflow has been refactored to use a custom MCP client that exposes vector search as an MCP tool. The MCP client now intelligently decides whether to invoke vector search, GraphDB query tools, or both, instead of always executing them in parallel.

The MCP client and GraphRAG agents intelligently route each query to the appropriate tool or tools instead of blindly calling all of them in parallel.

Simple RAG queries (e.g., "What does the regulation say about X?") → only Vector Search is called, saving 20–30 seconds of unnecessary MCP/GraphDB processing
Aggregate/structured queries (e.g., "How many articles were posted last October?") → only GraphDB tools are called, skipping unnecessary vector search
Complex queries that need both → both tools are still called, same as before

This significantly reduces latency and token usage for simple RAG queries that require only vector search, and improves handling of aggregate queries that require only GraphDB. The previous approach always triggered both paths, adding unnecessary MCP processing for straightforward queries.

More Robust GraphDB MCP Integration

Improved timeout handling and MCP prompt refinements result in more stable graph-tool calls in production workflows. MCP has been tested with different prompts, and improvements have been applied. Enterprise deployments benefit from increased reliability during graph-tool interactions, reducing workflow failures during demos and production usage.

In this section: