Kryon : AI Code Review

System Architecture

The system is built on a modern, event-driven architecture that separates code analysis from the web frontend.

GitHub
   │
   ├── Push / Pull Request Events
   ▼
Webhook Handler (Next.js)
   │
   ▼
Inngest Event Bus
   │
   ├── Repository Indexing Pipeline
   │      ├── Clone Repository
   │      ├── AST Chunking
   │      ├── Generate Embeddings (Gemini)
   │      └── Store Chunks + Vectors (Postgres + pgvector)
   │
   └── Pull Request Review Pipeline
          ├── Fetch PR Diff
          ├── Generate Embeddings
          ├── Retrieve Relevant Context
          ├── Generate AI Review (Gemini)
          ├── Post Review Comment to GitHub
          └── Store Review Results

Dashboard (Next.js)
   │
   ▼
PostgreSQL + pgvector

Core Pillars

AI Engine

Utilizes Gemini 2.0 Flash for deep architectural insights and automated PR reviews.

Background Processing

Powered by Inngest, ensuring reliable, retriable execution of code indexing and analysis.

Vector Search

Uses pgvector to retrieve relevant code context, enabling RAG (Retrieval-Augmented Generation) for code.

AST-Based Chunking

Uses Tree-sitter for intelligent code splitting, maintaining semantic context of functions and classes.

Explore the Architecture Learn how we process millions of lines of code with Inngest and pgvector.