Auto-Types

Understanding the 8 knowledge structure types: what they are and when to use them.

---

What Are Auto-Types?

Auto-Types are intelligent data structures that define how extracted knowledge is organized. Think of them as "containers" that shape what comes out of your documents.

Key characteristics:

Characteristic	Description
Type-Safe	Pydantic-based validation ensures consistent data
Self-Contained	Built-in operations (search, visualize, save)
Serializable	Save to disk, load later
Composable	Merge, update, extend incrementally

---

The 8 Auto-Types

Organized into two categories:

graph TD
    A[Auto-Types] --> B[Record Types<br/>Store data, no entity relations]
    A --> C[Graph Types<br/>Entity relationships]

    B --> B1[AutoModel<br/>One structured object]
    B --> B2[AutoList<br/>Ordered items]
    B --> B3[AutoSet<br/>Unique items]

    C --> C1[AutoGraph<br/>Binary relationships]
    C --> C2[AutoHypergraph<br/>Multi-entity relationships]
    C --> C3[AutoTemporalGraph<br/>+ Time]
    C --> C4[AutoSpatialGraph<br/>+ Location]
    C --> C5[AutoSpatioTemporalGraph<br/>+ Time + Location]

---

Record Types

For recording and storing data without relationships between entities.

Record types extract independent data items, where each item does not contain information about its relationship to other items.

AutoModel

Purpose: Extract a single structured object

Think of it as: A form with fields you want to fill in

Example use cases:

• Company earnings report (revenue, profit, growth)
• Product specification (name, price, features)
• Person profile (name, birth date, occupation)

Example output structure:

{
    "company_name": "Tesla Inc",
    "revenue": 81.46,
    "employees": 127855
}

Common templates: finance/earnings_summary, general/model

---

AutoList

Purpose: Extract an ordered collection

Think of it as: A ranked list or sequence

Example use cases:

• Top 10 movies of all time
• Step-by-step procedures
• Timeline events (simple chronological list)

Example output structure:

{
    "items": [
        {"name": "Tesla Coil", "year": 1891},
        {"name": "Radio", "year": 1898}
    ]
}

Key property: Order matters - items are in sequence

Common templates: general/list, legal/compliance_list

---

AutoSet

Purpose: Extract unique items (no duplicates)

Think of it as: A bag of unique tags

Example use cases:

• Keywords from a document
• Categories or tags
• Skill sets

Example output structure:

{
    "items": ["Electrical Engineering", "Physics", "Invention"]
}

Key property: Duplicates are automatically removed

Common templates: general/set, finance/risk_factor_set

---

Graph Types

For representing relationships between entities. Graph types include both the entities themselves (nodes) and the connections between them (edges).

Graph types are divided into subtypes based on relationship complexity and additional dimensional information:

Basic Graphs

AutoGraph

Purpose: Extract binary relationships (two entities at a time)

Think of it as: A social network diagram

Example use cases:

• People and their connections (worked with, married to)
• Concepts and their relationships (is-a, part-of)
• Organizations and interactions

Example output structure:

{
    "nodes": [
        {"name": "Tesla", "type": "person"},
        {"name": "AC Motor", "type": "invention"}
    ],
    "edges": [
        {"source": "Tesla", "target": "AC Motor", "type": "invented"}
    ]
}

Key property: Every relationship connects exactly two entities

Common templates: general/graph, general/biography_graph

---

AutoHypergraph

Purpose: Extract relationships involving 3+ entities

Think of it as: A project team where multiple people collaborate

Example use cases:

• Multi-party collaborations
• Complex interactions (buyer, seller, broker)
• Group memberships

Example output structure:

{
    "nodes": [...],
    "edges": [
        {
            "entities": ["Tesla", "Westinghouse", "Niagara"],
            "type": "collaboration"
        }
    ]
}

Key property: One "edge" can connect multiple entities

Common template: general/base_hypergraph

---

Enhanced Graphs (with dimensional information)

Built on basic graphs with added time or space dimensions for richer relationship descriptions.

AutoTemporalGraph (Temporal Graph)

Purpose: Extract relationships with time information

Think of it as: A timeline with connections

Example use cases:

• Biographies (life events with dates)
• Project histories
• Historical analysis

Example output structure:

{
    "edges": [
        {
            "source": "Tesla",
            "target": "AC Motor",
            "type": "invented",
            "time": "1888"
        }
    ]
}

Key property: Every relationship has a time component

Common templates: general/base_temporal_graph, finance/event_timeline

---

AutoSpatialGraph (Spatial Graph)

Purpose: Extract relationships with location information

Think of it as: A map with connections

Example use cases:

• Travel logs
• Geographic networks
• Location-based asset tracking

Example output structure:

{
    "entities": [
        {"name": "Colorado Springs", "location": "38.83,-104.82"}
    ],
    "relations": [
        {
            "source": "Tesla",
            "target": "Colorado Springs",
            "type": "conducted_experiments"
        }
    ]
}

Key property: Entities or relationships have geographic coordinates

Common template: general/base_spatial_graph

---

AutoSpatioTemporalGraph (Spatio-Temporal Graph)

Purpose: Extract relationships with both time and location

Think of it as: A historical map with dated events

Example use cases:

• Military history (battles with when and where)
• Epidemic tracking
• Historical migrations

Example output structure:

{
    "relations": [
        {
            "source": "Tesla",
            "target": "AC Motor",
            "type": "demonstrated",
            "time": "1888",
            "location": "Pittsburgh"
        }
    ]
}

Key property: Complete context - who, what, when, and where

Common template: general/base_spatio_temporal_graph

---

Selection Guide

Decision Tree

What do you need to extract?
│
├─ Single structured object (like a form)
│  └─ AutoModel
│
├─ Collection of items
│  ├─ Order matters? → AutoList
│  └─ Just unique items? → AutoSet
│
└─ Relationships between entities → Graph Types
   ├─ Relationship arity
   │  ├─ Binary (2 entities) → Basic Graphs
   │  │  ├─ Need time info? → AutoTemporalGraph
   │  │  ├─ Need location info? → AutoSpatialGraph
   │  │  ├─ Need both? → AutoSpatioTemporalGraph
   │  │  └─ Neither? → AutoGraph
   │  │
   │  └─ Multi-entity (3+ entities) → AutoHypergraph

By Use Case

Use Case	Auto-Type	Why
Company financial report	AutoModel	Single structured summary
Top 10 list	AutoList	Ordered/ranked items
Keywords/tags	AutoSet	Unique collection
People network	AutoGraph	Binary relationships
Project teams	AutoHypergraph	Multi-person collaborations
Biography timeline	AutoTemporalGraph	Events with dates
Travel itinerary	AutoSpatialGraph	Places with connections
Historical events	AutoSpatioTemporalGraph	Full context needed

Quick Reference

Record Types

Auto-Type	Output Shape	Key Feature
AutoModel	Single object	Fixed schema fields
AutoList	Array	Maintains order
AutoSet	Array	Deduplicates

Graph Types

Auto-Type	Output Shape	Key Feature
AutoGraph	Nodes + Edges	Binary relations (2 entities)
AutoHypergraph	Nodes + Hyperedges	Multi-entity relations (3+ entities)
AutoTemporalGraph	Nodes + Time-Edges	Binary relations + time dimension
AutoSpatialGraph	Nodes + Geo-Edges	Binary relations + geographic dimension
AutoSpatioTemporalGraph	Nodes + Spatio-Temporal-Edges	Binary relations + time + space

---

Next Steps

Ready to use Auto-Types? Choose your path:

Quick Start: Use built-in Templates — no Auto-Type knowledge needed

More Control: Learn how to work with Auto-Types in Python: - Working with Auto-Types — practical examples - Python SDK Quickstart — get started in 5 minutes

Advanced: Create custom Auto-Type configurations - Creating Custom Templates - API Reference