Skip to main content

Architecture

ConCRG embeds into your application as a lightweight connector and communicates with two backend services that handle knowledge extraction and AI querying.

System Overview​

CRG System Architecture

Components​

CRG Connector​

The connector is a client-side library that embeds in your application. It:

  • Observes navigation and interface changes
  • Captures page state during training
  • Renders the training panel and assist toolbar in an isolated layer (no CSS conflicts with your app)
  • Communicates with the Train Service during training and the Graph Service during assist

Train Service​

A lightweight backend that handles all knowledge extraction during training:

  • Analyzes interface snapshots using AI
  • Processes frontend source code to extract routes, models, and permissions
  • Crawls documentation sites and extracts knowledge
  • Accepts conversational input from your team
  • Persists extracted knowledge as structured facts

Key Design Decisions​

Why an isolated UI layer? ConCRG's UI renders in a fully isolated container so its styles never interfere with your application — and your application's styles never affect ConCRG.

Why file-based knowledge store by default? The default knowledge store works with zero infrastructure — no database required to get started. Teams can connect a persistent graph store when they need advanced querying at scale.