Entities & Relationships

This page explains how entities and relationships are extracted, stored, and used in AgentisPro RAG Engine.

Entity Extraction Process

When you upload a document, the following happens:

Document Text
      │
      ▼
┌─────────────┐
│   LightRAG  │ ← NLP Processing
│  Extraction │
└──────┬──────┘
       │
       ├──────────────────────────────────────┐
       ▼                                      ▼
┌─────────────┐                        ┌─────────────┐
│  Entities   │                        │Relationships│
│  Detected   │                        │  Detected   │
└──────┬──────┘                        └──────┬──────┘
       │                                      │
       └──────────────────┬───────────────────┘
                          ▼
                   ┌─────────────┐
                   │    Neo4j    │
                   │   Storage   │
                   └─────────────┘

Entity Types

Person 👤

Individuals mentioned in documents.

Extraction triggers:

• Proper nouns followed by roles ("John Smith, CEO")
• Email patterns
• Signature blocks
• Author metadata

Organization 🏢

Companies, departments, teams.

Extraction triggers:

• Legal suffixes (Inc., LLC, Corp.)
• Department keywords ("Sales Team", "HR Department")
• Domain names in emails

Product 📦

Products, services, features.

Extraction triggers:

• Product naming patterns
• Version numbers
• SKU codes
• Trademark symbols (™, ®)

Location 📍

Physical and virtual places.

Extraction triggers:

• Address patterns
• City/country names
• Building names
• Geographic references

Event 📅

Meetings, deadlines, milestones.

Extraction triggers:

• Date patterns
• Event keywords ("meeting", "conference", "deadline")
• Calendar references

Concept 💡

Abstract ideas, topics, categories.

Extraction triggers:

• Repeated themes across documents
• Section headers
• Tag-like patterns

Relationship Types

Common Relationships

Relationship Properties

Relationships can have properties:

// Example: Weighted relationship
(john:Person)-[:WORKS_AT {since: "2020", role: "CEO"}]->(acme:Organization)

Entity Linking

When the same entity appears in multiple documents, AgentisPro links them:

Document 1: "John Smith, CEO of Acme"
Document 2: "J. Smith presented the quarterly results"
Document 3: "CEO John Smith announced..."

→ All linked to single entity: John Smith (Person)

Linking Algorithm

1. Exact match: Same name → 100% match
2. Fuzzy match: Similar names → Calculate similarity
3. Context match: Same relationships → Increase confidence
4. User override: Manual linking/unlinking

Creating Entities Manually

Via Web UI

1. Navigate to Knowledge Graph
2. Click + Add Entity
3. Fill in properties:
   • Name (required)
   • Type (required)
   • Custom properties
4. Save

Via API

POST /knowledge-graph/entities
Content-Type: application/json
 
{
  "name": "New Customer",
  "type": "Organization",
  "properties": {
    "industry": "Healthcare",
    "size": "Enterprise"
  }
}

Creating Relationships Manually

Via Web UI

1. Select source entity
2. Click + Add Relationship
3. Search for target entity
4. Select relationship type
5. Add properties (optional)
6. Save

Via API

POST /knowledge-graph/relationships
Content-Type: application/json
 
{
  "source_id": "entity_123",
  "target_id": "entity_456",
  "type": "works_at",
  "properties": {
    "since": "2025-01-01"
  }
}

Editing Entities

Merge Duplicates

If the same entity was created twice:

1. Select both entities (Ctrl+Click)
2. Right-click → Merge Entities
3. Choose primary entity (keeps properties)
4. Confirm

Split Entity

If different entities were incorrectly merged:

1. Select entity
2. Click Edit → Split
3. Define which relationships go to each new entity
4. Confirm

Best Practices

Naming Conventions

Relationship Consistency

Use consistent relationship types:

• ✅ works_at (not employed_by, works_for)
• ✅ authored (not wrote, created)
• ✅ manages (not supervises, leads)

Quality Control

Regular maintenance:

1. Review new entities weekly
2. Merge duplicates
3. Delete false positives
4. Add missing relationships

Querying Entities

Find All of Type

GET /knowledge-graph/entities?type=Person

Find by Name

GET /knowledge-graph/entities?search=John

Find Related

GET /knowledge-graph/entities/{id}/related?depth=2

Next Steps