AI Software Team

Autonomous multi-agent engineering runtime built with LangGraph

README.md

Introduction

This project simulates a real AI engineering organization where specialized agents collaborate to analyze repositories, retrieve code context, modify implementations, validate changes, and stream runtime activity live to the UI.

🤖 Multi-Agent Architecture

Intent ClassifierStrategic PlannerRepository SearcherTask ClassifierBackend EngineerFrontend EngineerReviewerValidatorChange SummaryAnswer Agent
Setup.zsh
âžœInstall Dependencies
npm install
âžœRun ChromaDB
docker run -p 8000:8000 chromadb/chroma
âžœLaunch Systems
npm run dev
Architecture.drawio

Runtime Pipeline

1
User Task
2
Intent Classifier
3
Planner
4
Searcher
5
Engineer
6
Validator
MemorySystem.ts

Persistent semantic memory learns repository architecture, conventions, and recurring failures.

Memory Retrieval Loop
1. Extraction
2. Embedding Gen
3. Vector Storage
4. Semantic Retrieval
Tools.json
Intelligence
search, read, grep
Engineering
patch, diff, terminal
Validation
build, typecheck, lint
TechStack.yml

Backend

  • • TypeScript / Node.js
  • • LangGraph Orchestration
  • • Hono Framework

Frontend

  • • Next.js / React
  • • ReactFlow Visualization
  • • Tailwind CSS
Events.ts
type RuntimeEvent =
  | "agent" | "tool"
  | "terminal" | "diff"
  | "approval" | "log";

Real-time streaming via Server-Sent Events (SSE)

Tasks.txt
Example Requests:
- Add auth request logging
- Fix Prisma validation issue
- Implement dark mode toggle
Future.log
  • • Durable Workflow Resumption
  • • Playwright Browser Agent
  • • AST-safe Code Editing
  • • Multi-Repo Intelligence
Security.md
Experimental system. Dangerous operations (rm, push) require Human Approval Gates. Always review generated code.
STAR REPOSITORY ON GITHUB

End of Documentation — Autonomous AI Engineering System