GraphRAG 1.3.0 Release Notes

GraphRAG Evaluation tool - AI Evaluation & Benchmarking
A self-contained evaluation framework that measures both retrieval and generation quality using industry-standard RAGAS metrics, runs against any GraphRAG deployment, and produces actionable HTML reports (an aggregated overview for release-level assessment and a per-question breakdown for debugging). It supports multiple LLMs, including OpenAI (Azure Foundry), OpenAI, Bedrock, Gemini, and Claude. Access the GitHub repo, including README and provider-specific run instructions for the .jar.
Added automated benchmark baselines for repeatable regression testing across releases, including per-scenario metrics, token usage, and execution time in the reports.
Full concept attribute support in the UI
Scope notes, broader/narrower concepts, related concepts, custom attributes, and custom relations from Graph Modeling, with clickable navigation between related concepts.
OAuth2 replaces Basic Auth for all Graph Modeling API integrations, resolving parallel request failures under Keycloak protection.
UI quality-of-life improvements
Collapsible subthreads in conversations, WCAG AA contrast for info, warning, and error messages, and refreshed input-field text.
Conversations and Components Services upgraded to Spring Boot 4.1.0 and Jackson 3
Hidden labels and concept scheme information now exposed in the API response.

Improved visibility of info, warning, and error states, meeting WCAG AA contrast standards.
You can now collapse subthreads within each conversation.
Added options for automated evaluation test reports with per-scenario metrics, token usage, and execution time.
Changed the MCP protocol to HTTP Streamable for n8n-to-GraphDB communication.
System prompts are now passed in the correct input fields in n8n AI Agent nodes.
LLMs can now get context for custom relations and concept attributes through the new configuration property conceptPropertyLlmContext.
Broader/narrower concepts are now clickable, and the details are fetched via the new /conceptDetails endpoint in the Conversations Service. The endpoint accepts an array of concept URIs and can also be used as a standalone webhook.

Added a new branded n8n image overlaid in GraphRAG.
Fixed vulnerabilities in spring-boot, HTTPCore5, and OpenSSL in the Conversations Service base image.
Updated the n8n base image to reduce the incidence of critical vulnerabilities.
The n8n API key has been moved from the data table to credentials and is no longer stored in plain text. A new
finalizeInit endpoint in the Conversations Service lets you update the credential after initialization.

Conversations API now rejects empty questions with proper validation.
Misspelled variables vectorStoreCustomMappings and parallelStepsMaxTimeout have been corrected.

Component	Versions
n8n / GraphRAG Workflow Engine	2.4.4 – 2.34.6
Keycloak	26.3.2
PostgreSQL	17
DuckDB JDBC	1.3.2.1
NGINX	1.29.0
Elasticsearch	8.5 – 9.3.3
OpenSearch — self-hosted	3.2 and later
Amazon OpenSearch Serverless	2.17.x and later
AWS Bedrock	Currently supported AWS Bedrock services and models (Compatibility depends on the LLM models configured for the deployment and their availability in AWS Bedrock.)
Azure OpenAI	Current Azure OpenAI Embeddings API (The Components Service uses com.openai:openai-java:4.16.1.)

The following artifacts were released alongside GraphRAG: