PRODUCT UPDATE

See Why Your AI Answered That Way: Introducing Subgraph Retrieval

Full transparency into the entities, relationships, and sources behind every response. Built for teams that need audit trails.

The Black Box Problem

You deploy a RAG system. It answers questions. But when someone asks "why did it say that?" — you're stuck.

Most RAG platforms give you a list of retrieved chunks. Maybe you see the top 5 documents that matched the query. But that's not enough. You don't know:

• Which specific facts from those documents influenced the answer?
• What relationships were traversed to connect the dots?
• Why these sources and not others?
• If the answer is wrong, where did the reasoning break down?

For internal tools, this is frustrating. For regulated industries — finance, healthcare, legal — it's a compliance risk. When decisions are informed by AI, you need to show your work.

Why Explainability Matters

Regulatory frameworks increasingly require AI systems to be explainable. If an AI-assisted decision is challenged — a loan denial, a medical recommendation, a legal assessment — you may need to reconstruct exactly what information led to that output.

Explainability enables:

• Audit compliance — Demonstrate to regulators exactly how AI-assisted decisions were made
• Error diagnosis — When an answer is wrong, pinpoint where the retrieval or reasoning failed
• User trust — End users can verify that answers are grounded in real sources, not hallucinations
• Continuous improvement — Understand retrieval patterns to optimize your knowledge base

The Solution: Retrieve the Subgraph

Our GraphRAG debugging tools now let you retrieve the exact subgraph used to generate any response. Not just "these documents matched" — but the complete reasoning path:

• Entities — Which people, organizations, products, or concepts were identified as relevant
• Relationships — What connections were traversed (e.g., "reports to", "subsidiary of", "referenced in")
• Source documents — The specific chunks that provided evidence, with citations
• Traversal path — How the system moved through the knowledge graph to gather context

The subgraph view shows exactly which entities and relationships contributed to the response.

How It Works

When you make a query to your GuidedMind agent, the system:

1. Identifies relevant entities in your query
2. Traverses the knowledge graph to find connected information
3. Retrieves supporting documents via hybrid BM25 + vector search
4. Generates a response grounded in the retrieved context
5. Records the subgraph — all entities, edges, and sources used

You can retrieve this subgraph at any time — through the dashboard UI or via API. It's your audit trail.

Use Cases

Compliance & Audit

A financial services firm uses GuidedMind to help analysts research investment opportunities. When a recommendation is questioned, they can pull the exact subgraph — showing which SEC filings, news articles, and internal reports informed the analysis.

Debugging & Quality Assurance

Your AI gives an incorrect answer. Instead of guessing what went wrong, you retrieve the subgraph and see: the relevant document existed, but a relationship edge was missing. You update the knowledge base. Problem solved.

User Trust & Transparency

End users can click "show sources" and see not just document titles, but the specific entities and connections that led to their answer. Transparency builds confidence in AI-assisted decisions.

Getting Started

Subgraph retrieval is available now for all accounts using GraphRAG.

• Dashboard: Click any response in your agent chat, then select "View Subgraph" to see the visual representation
• API: Pass include_subgraph: true in your request to receive the subgraph data in the response payload

The subgraph data includes entity IDs, relationship types, source document references, and relevance scores — everything you need for audit logs or custom visualizations.

Try it now