GraphRAG 1.0.0-RC1 Release Notes

Hybrid Retrieval (Graph + Vector + Full-Text Search):

GraphRAG achieves superior accuracy for complex, multi-hop queries by combining three powerful retrieval methods - relation-based queries via SPARQL across knowledge graphs; semantic similarity employing the vector search; and keyword matching utilizing full-text search.

This approach finds relevant information that single-method systems miss, especially for complex questions requiring multiple data connections.

Context-Aware Query Enhancement

User questions are automatically enriched using your organization's taxonomy and ontologies. The system automatically expands brief or unclear queries with relevant context, ensuring the AI understands what you are actually asking for.

Auditable, Transparent Explainability

Every answer is fully transparent. Users receive provenance metadata, citations, and clickable semantic concepts, allowing them to inspect the exact source documents and graph entities used. This level of detail is crucial for auditability and meeting regulatory compliance requirements.

Detailed Source Tracing

A dedicated panel provides structured information about the source materials including document metadata, and URLs.

Transparent Agent Decision Logic

An Explainability Panel is available to detail the agent's decision-making process. This view exposes the retrieval steps and the specific tools that were utilized to arrive at the answer.

Integrated Security & Compliance Guardrails

Embedded native guardrails actively filter input and output content ensuring adherence to corporate security standards by rejecting any inappropriate, off-topic, or non-compliant content before it reaches the user.

Vendor-Agnostic Infrastructure:

Decoupled architecture supports major LLM providers (OpenAI, Azure AI, Bedrock), various vector databases, and different Identity Providers. This prevents any vendor lock-in and lets IT change models or infrastructure as technology evolves.

Agentic Orchestration via n8n:

n8n acts as the high-level orchestrator for the retrieval workflow. Dedicated sub-workflows (e.g., concept extraction, answer enrichment) enable rapid experimentation, chaining of components, and custom enterprise flows - all without modifying the core codebase.

Multi-turn conversations with context retention

Suggested follow-up questions based on current discussion facilitating a deeper and more structured exploration of the data

All user sessions and conversation histories are securely stored, allowing users to easily revisit and reference past interactions

Real-time streaming responses via Server-Sent Events (SSE) with formatted content

SKOS Concept Highlighting

When responses reference concepts from the Knowledge Graph, those terms are automatically highlighted in the response. Users can click on any highlighted concept to access the SKOS properties including definitions (skos:definition), synonyms (skos:altLabel), and notes (skos:scopeNote) retrieved through the extraction API.

Authentication & Roles

Secure user authentication is managed via Keycloak (OIDC), supporting standard user roles and session management without requiring complex SSO configurations in the current release.

Multi-Stage Guardrails

Guardrails monitor both input and output stages across all workflows, ensuring compliance and user protection.

Sync & Async Querying

The API provides both synchronous (immediate response) and asynchronous (polling/streaming) modes, allowing you to handle long-running queries without freezing your application interface.

Client & Browser Persistence

Persistent client IDs maintain seamless sessions across browser refreshes and multiple tabs.

Cross-Language Content Processing

The system supports multilingual knowledge bases with content processing and querying capabilities across multiple languages (currently primarily validated in English).

Graphwise Platform: PoolParty 10.0+, GraphDB 11.1+

Orchestrator: n8n 1.121.3

Java Runtime: OpenJDK 21

NodeJS >=18