BACK TO HOME

AutoSwe

An AI-powered automation system for processing Git repositories, generating PRs, and maintaining vectorized repo metadata for advanced analysis.

Technical Architecture

High-level architecture

AI Engineer Workflow

Features

  • Repo Indexing: Automatically indexes repository metadata into PGVector. Skips indexing if already present.
  • Automated PR Generation: Processes repositories using LangGraph nodes. Creates Pull Requests based on AI-generated analysis or updates.
  • Incremental Updates: Watches push events to main or master branches. Updates indexed data incrementally without reprocessing the whole repo.
  • Queue & Worker System: Uses BullMQ for background job management. Supports scalable processing with multiple workers.
  • Sandboxed Execution: Runs code safely using E2B sandbox.

Tech Stack

  • Language & Runtime: Bun
  • Monorepo Management: Bun workspace
  • Database: PostgreSQL + PGVector for embeddings
  • Queue System: BullMQ
  • Sandboxing: E2B
  • AI Processing: LangGraph nodes for repo analysis