> ## Documentation Index
> Fetch the complete documentation index at: https://docs.mindblix.tech/llms.txt
> Use this file to discover all available pages before exploring further.

# Knowledge Graph

> The Structured Knowledge Graph is MINDBLIX's memory system — enabling grounded, hallucination-free AI reasoning.

## Overview

The **Structured Knowledge Graph** is a persistent, evolving representation of your organization's information. It stores entities, relationships, and facts with confidence scores, enabling agents to produce grounded, verifiable outputs.

> [!NOTE]
>
> The Knowledge Graph is what makes MINDBLIX different from standard RAG (Retrieval-Augmented Generation) systems. Instead of just retrieving text chunks, agents reason over a structured graph of verified facts and relationships.

***

## How It Works

```mermaid theme={null}
graph TD
    A[Data Sources] --> B[Entity Extraction]
    B --> C[Relationship Mapping]
    C --> D[Confidence Scoring]
    D --> E[Knowledge Graph]
    E --> F[Agent Queries]
    F --> G[Grounded Responses]

    style E fill:#9300FF,color:#fff
```

### Data Flow

1. **Ingestion** — Raw data arrives from connected sources (docs, APIs, databases)
2. **Extraction** — Entities (people, orgs, concepts) are automatically identified
3. **Linking** — Relationships between entities are mapped and verified
4. **Scoring** — Each fact receives a confidence score based on source reliability
5. **Querying** — Agents query the graph during reasoning for factual grounding

***

## Entity Types

| Entity | Description | Example |
| - | - | - |
| **Person** | Individual people | `John Smith, CEO` |
| **Organization** | Companies, teams | `Acme Corp, Engineering Team` |
| **Concept** | Abstract ideas, technologies | `Machine Learning, REST API` |
| **Document** | Source documents | `Q4 Report 2025.pdf` |
| **Event** | Timestamped occurrences | `Product Launch, March 2026` |
| **Metric** | Quantifiable values | `Revenue: $2.4M, ARR` |

***

## Confidence Scoring

Every fact in the knowledge graph has a confidence score:

| Score Range | Level | Meaning |
| - | - | - |
| 0.95 – 1.00 | **Verified** | Multiple reliable sources confirm |
| 0.80 – 0.94 | **High** | Strong single-source evidence |
| 0.60 – 0.79 | **Medium** | Some supporting evidence |
| 0.40 – 0.59 | **Low** | Limited evidence, may need verification |
| 0.00 – 0.39 | **Unverified** | Flagged for human review |

> [!WARNING]
>
> Agents will not use facts with a confidence score below the configured `min_confidence` threshold (default: 0.80) in their outputs.

***

## Temporal Awareness

The knowledge graph tracks **when** information was added and whether it has been superseded:

```json theme={null}
{
  "entity": "Acme Corp",
  "attribute": "CEO",
  "value": "Jane Doe",
  "valid_from": "2025-06-01",
  "valid_to": null,
  "previous": {
    "value": "John Smith",
    "valid_from": "2020-01-15",
    "valid_to": "2025-05-31"
  }
}
```

This enables agents to answer time-sensitive questions accurately.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.