Skip to main content
Superatom is built as a modular, multi-layer architecture designed for enterprise scale, security, and flexibility.

High-Level Architecture


Layer Overview

Presentation Layer

The user-facing components:

Infrastructure Layer

Handles real-time communication and scaling:

Intelligence Layer

The “brain” of the platform:

Data Layer

Persistent storage:

Request Flow


Agent-Based Architecture

Superatom uses an agent-based architecture where each data source gets its own dedicated, containerized agent. The Main Agent (orchestrator) coordinates everything but never directly accesses customer databases or holds database credentials. This separation provides security isolation (the LLM-facing component cannot access data), independent scaling per data source, and fault isolation (one failing agent doesn’t affect others).

Agent Architecture

Deep dive into agent isolation, communication, and scaling

Deployment Models

Superatom supports two deployment configurations:

Cloud-Hosted

Uses Cloudflare Workers and Durable Objects for the infrastructure layer. WebSocket connections are managed at the edge with hibernation support for cost efficiency.

On-Premise (Enterprise)

All components, including the WebSocket server, Main Agent, data-source agents, and databases, run on the customer’s infrastructure as Docker containers orchestrated with Docker Compose or Kubernetes. No external dependencies required (with self-hosted LLM option).

Infrastructure

Deployment specs, scaling, and on-premise requirements

Key Design Decisions

WebSocket-First Communication

All real-time communication uses WebSockets:

Bi-directional

Server can push updates without polling

Streaming

Responses appear as they’re generated

Efficient

Lower latency than REST for interactive apps

Hibernatable

90-99% cost reduction with Cloudflare DO

Multi-LLM Support

No vendor lock-in:

Type Safety Throughout

Full TypeScript with Zod validation:
  • Compile-time type checking
  • Runtime message validation
  • Schema generation for API documentation

Scalability

Horizontal Scaling

Each project has:
  • Dedicated Durable Object for WebSocket handling
  • Independent scaling based on load
  • Complete isolation from other projects

Cost Efficiency


Technology Stack

Frontend

Backend

Infrastructure

DevOps


Next Steps

Agent Architecture

Agent isolation, communication, and scaling

Data Flow

Detailed request/response flow

AI Engine

How the intelligence layer works

Modules

Deep dive into each module

Infrastructure

Deployment specs and scaling