BACK TO HOME

CodeDaddy

An AI-powered PR review workflow that processes commits asynchronously using a multi-agent architecture for deep code analysis inspired by code-rabbit.

Technical Architecture

High-level architecture

The following diagram illustrates the high-level architecture of the CodeDaddy PR Review Workflow, showing how pull requests and commits are processed asynchronously and analyzed by multiple AI agents in parallel.

Architecture Diagram

System Design

Overview

  • Webhook Trigger — When a Pull Request or Commit event occurs, a webhook receives the event payload.
  • Background Processing — The event is pushed into a Python RQ (Redis Queue) background job (GITHUB_PR_PROCESS).
  • Workers — Multiple workers clone the repository, parse source files, build semantic graphs, and prepare contextual data. Each worker uploads its processed context to S3.
  • Context Initialization — A controller process downloads context from S3 and sets up the state for LangGraph nodes.
  • Parallel Agent Execution — Several agents run concurrently:
    • code_quality_agent
    • performance_agent
    • security_agent
    • test_agent
  • Aggregation Phase — The aggregator_agent combines insights from all agents.
  • Comment Update — The aggregated output replaces the initial loading comment with a comprehensive final PR review comment.

Key Characteristics

  • Event-driven: Triggered by GitHub webhooks.
  • Asynchronous: Uses background workers for parallel processing.
  • Scalable: Each worker operates independently.
  • AI-powered: Multi-agent architecture performs different aspects of code analysis.
  • Automated Feedback: Automatically updates PR comments with contextual insights.