Core Concepts

Before diving deeper, understand these core concepts that power Superatom.

The Decision Loop

Superatom implements a continuous decision-making loop:

Stage	What Happens
Understand	Connect data, build semantic model
Analyze	AI finds patterns, anomalies, insights
Suggest	Recommend actions to address findings
Verify	Human reviews and approves
Execute	Action committed to systems
Observe	Track outcomes, learn from results

Semantic Model

The semantic model is Superatom’s understanding of your data:

Schema Understanding

Tables, columns, relationships, data types

Statistical Profile

Distributions, patterns, common values

Domain Context

Industry-specific interpretation

Query Capability

What questions can be asked

The semantic model enables natural language queries by mapping human questions to database operations.

Tribal Knowledge

Organization-specific knowledge captured at three levels:

Level	Example	Scope
Global	”Sales exclude returns”	All users, all queries
User	”Show only West Coast data”	Specific user
Query	”Deadstock = no sales in 12 months”	Specific question type

Generative UI

Automatic visualization creation:

User Asks Question

Natural language query submitted

Data Retrieved

SQL generated and executed

Data Analyzed

Result set profiled for characteristics

Component Selected

AI chooses optimal visualization

UI Rendered

Interactive component displayed

16+ component types automatically selected based on data patterns.

Knowledge Nodes

Building blocks of organizational knowledge:

knowledge_node:
  title: "Overstock Definition"
  type: query

  description: |
    Items where inventory exceeds 90 days
    of average sales velocity

  rules:
    - "Calculate days_supply = qty / avg_daily_sales"
    - "Overstock when days_supply > 90"

  applies_to:
    - "inventory queries"
    - "overstock analysis"

Knowledge nodes guide how the AI interprets and answers questions.

Knowledge Graph

Visual representation of how concepts connect:

The graph shows relationships between:

Entities - Orders, Customers, Products
Processes - Workflows, lifecycles
Insights - Derived knowledge, calculations

Coding Agents

AI agents that understand business context: Agents can:

Break complex questions into sub-tasks
Generate and execute multiple queries
Iterate until they find the answer
Explain their reasoning

Dry-Run Execution

Safe action execution with human oversight: Actions are simulated first, showing what WOULD happen before committing.

Project Isolation

Each deployment is completely isolated:

Separate Infrastructure

Dedicated Durable Objects per project

Data Isolation

No cross-project data access

Independent Scaling

Resources scale per project needs

Security Boundary

Complete security separation

Real-Time Streaming

Responses appear as they’re generated:

Text streams character by character
Charts build progressively
Tables populate row by row
Follow-ups appear after main content

This creates a more engaging, responsive experience than waiting for complete results.

Multi-LLM Support

No single AI vendor dependency:

Provider	When Used
Anthropic Claude	Complex reasoning tasks
OpenAI GPT	General purpose queries
Google Gemini	Fast processing needs
Groq	Real-time, high-volume

Switch providers based on task, cost, or preference.

Key Terminology

Term	Definition
Semantic Model	AI’s understanding of your data structure and meaning
Tribal Knowledge	Organization-specific context captured in knowledge nodes
Knowledge Node	Individual piece of organizational knowledge
Knowledge Graph	Visual map of how concepts relate
Generative UI	Automatic visualization creation
Coding Agent	AI that iteratively solves complex problems
Dry-Run	Simulated execution before commitment
Durable Object	Cloudflare’s stateful serverless unit

Getting Started

Product

Architecture

Security & Data Privacy

Intellectual Property

Implementation

Use Cases

Case Studies

API Reference

Guides

The Decision Loop

Semantic Model

Schema Understanding

Statistical Profile

Domain Context

Query Capability

Tribal Knowledge

Generative UI

Knowledge Nodes

Knowledge Graph

Coding Agents

Dry-Run Execution

Project Isolation

Separate Infrastructure

Data Isolation

Independent Scaling

Security Boundary

Real-Time Streaming

Multi-LLM Support

Key Terminology

Next Steps

Quickstart

Architecture

​The Decision Loop

​Semantic Model

Schema Understanding

Statistical Profile

Domain Context

Query Capability

​Tribal Knowledge

​Generative UI

​Knowledge Nodes

​Knowledge Graph

​Coding Agents

​Dry-Run Execution

​Project Isolation

Separate Infrastructure

Data Isolation

Independent Scaling

Security Boundary

​Real-Time Streaming

​Multi-LLM Support

​Key Terminology

​Next Steps

Quickstart

Architecture

The Decision Loop

Semantic Model

Tribal Knowledge

Generative UI

Knowledge Nodes

Knowledge Graph

Coding Agents

Dry-Run Execution

Project Isolation

Real-Time Streaming

Multi-LLM Support

Key Terminology

Next Steps