🩹 AutoHeal 2.0 — AI-Powered Self-Healing CI/CD Platform

AutoHeal is a full-stack SaaS platform that automatically detects CI/CD failures, diagnoses root causes using AI, and generates pull requests with fixes — all without human intervention.

🖼️ Product Showcase

The Command Center: A high-fidelity, interactive hub for autonomous CI

Seamless Control: Manage repository connections and monitor live healing pipelines in real-time via Server-Sent Events (SSE).

---

🏗️ Architecture & Logic Flow

AutoHeal 2.0 operates on a weightless, event-driven architecture designed for extreme scalability and cost-efficiency.

graph TD
    subgraph "GitHub Ecosystem"
        GH_Repo[("Target Repository")]
        GH_Actions["GitHub Actions (CI)"]
        GH_Webhooks["GitHub Webhooks"]
    end

    subgraph "AutoHeal 2.0 (Backend)"
        API_Gateway["Express API Server"]
        Webhook_Handler["Webhook Handler & Mutex"]
        Ingestor["Background Ingestion Engine"]
        
        subgraph "AI Core"
            RAG["RAG Retrieval (Atlas)"]
            AST["AST Parser (Tree-sitter)"]
            RCA["AI Root Cause Analysis"]
            Fixer["AI Selective Patching"]
        end
        
        Git_Ops["Git Operations Hub"]
    end

    subgraph "Data & AI Layer"
        MongoDB[("MongoDB Atlas + Vector Search")]
        Groq["Groq (Llama 3.3 70B)"]
        Gemini["Google Gemini (Fallback)"]
        Xenova["Local Embeddings (WASM)"]
    end

    subgraph "Frontend"
        Dashboard["React Dashboard (SSE Updates)"]
    end

    GH_Repo -- "Push/Failure" --> GH_Actions
    GH_Actions -- "Event: workflow_run" --> GH_Webhooks
    GH_Webhooks -- "Secured Payload" --> Webhook_Handler
    Webhook_Handler -- "Retrieve Context" --> RAG
    RAG -- "Vector Search" --> MongoDB
    Webhook_Handler -- "Run RCA" --> RCA
    RCA -- "Request Files" --> Git_Ops
    Git_Ops -- "Fetch Source" --> GH_Repo
    RCA -- "Strategy" --> Fixer
    Fixer -- "Generate JSON Patch" --> Groq
    Groq -- "Fallback" --> Gemini
    Fixer -- "Patch in Memory" --> Git_Ops
    Git_Ops -- "Shadow Push" --> GH_Repo
    GH_Repo -- "CI Result" --> Webhook_Handler
    Webhook_Handler -- "Success" --> Dashboard
    Webhook_Handler -- "Failure" --> RCA

Loading

1. 👻 Shadow Branching (The Zero-Infra Testing)

Instead of running heavy CI tests locally, AutoHeal offloads 100% of the testing compute infrastructure to GitHub.

• After generating a fix, the backend pushes code to a hidden, temporary branch (fix/autoheal-...).
• GitHub Actions runs the user's native CI suite on this branch.
• AutoHeal intercepts the result via webhooks. If it passes, only then is a PR opened. This ensures developers never see a failing AI PR.

2. 🧠 Local RAG & AST Chunking

To scale to massive codebases without hitting AI token limits, AutoHeal utilizes Retrieval-Augmented Generation:

• Background Ingestion: When a repo is enabled, a non-blocking worker clones it, parses it via web-tree-sitter (AST), and extracts individual functions.
• Local Embeddings: Code chunks are converted into vectors locally using Xenova Transformers (all-MiniLM-L6-v2) running in WASM—zero API cost.
• Vector Search: During a crash, AutoHeal performs a $vectorSearch in MongoDB Atlas to surgically find the exact function responsible for the error.

3. ♻️ Smart Retry Loops

If an AI-generated fix fails validation on the Shadow Branch, the system doesn't give up. It extracts the new error logs, feeds the previous attempt back into the AI as a "failed lesson," and triggers a recursive healing cycle until the code passes all tests.

---

✨ Key Features

• 🔐 Enterprise Security — GitHub tokens encrypted with AES-256-GCM; HMAC-signed webhooks.
• 🤖 Multi-Model Fallback — Primary RCA on Groq (Llama 3.3 70B) with instant cascade to Google Gemini 2.0 Flash if rate limits are hit.
• ⚡ Real-Time SSE — Live pipeline status updates pushed to the frontend via Server-Sent Events—no polling needed.
• 🛠️ JSON Selective Patching — AI outputs search/replace blocks instead of full files, slashing token consumption by 90% and preventing indentation errors.
• 🚫 Infinite Loop Protection — Built-in circuit breakers limit executions per repository to prevent runaway CI costs.

---

📋 Tech Stack

Layer	Technology
Frontend	React 19, Vite 6, Framer Motion, Tailwind CSS, Lucide React
Backend	Node.js, Express, Passport.js (GitHub OAuth), JWT
Database	MongoDB Atlas (Vector Search + Aggregation Pipelines)
AI (RCA/Fix)	Groq (Llama-3.3-70b-versatile), Google Gemini 2.0 Flash
AI (Embeddings)	Xenova Transformers (WASM), all-MiniLM-L6-v2
Parsing	web-tree-sitter (Abstract Syntax Tree analysis)
Real-time	Server-Sent Events (SSE)

---

🛠️ Setup

Prerequisites

• Node.js 18+
• MongoDB Atlas account (with a Vector Search Index named vector_index)
• GitHub OAuth App
• Groq API key & Gemini API key
• Publicly accessible Backend URL (like Render, or ngrok for local dev)

1. Installation

git clone https://github.com/CodeByAfroj/AutoHeal.git
cd AutoHeal

# Install Backend
cd backend && npm install

# Install Frontend
cd ../frontend && npm install

2. Environment Configuration

Create backend/.env:

GITHUB_CLIENT_ID=your_id
GITHUB_CLIENT_SECRET=your_secret
# AI Providers
GROQ_API_KEY=your_key
GEMINI_API_KEY=your_key

# Database
MONGODB_URI=your_atlas_uri

# GitHub App Integration
GITHUB_CLIENT_ID=...
GITHUB_CLIENT_SECRET=...
WEBHOOK_SECRET=...

# Security
JWT_SECRET=...
ENCRYPTION_KEY=... (32-character string for AES-256)

2. Global Vector Index

Create a Search Index named vector_index on the codechunks collection:

{
  "fields": [
    { "numDimensions": 384, "path": "embedding", "similarity": "cosine", "type": "vector" },
    { "path": "repositoryId", "type": "filter" }
  ]
}

---

📁 Project Structure

AutoHeal/
├── backend/
│   ├── routes/
│   │   ├── webhook.js         # Core Event Logic & Shadow Branching
│   │   ├── repos.js           # Repo Mgmt & Auto-Webhook Creation
│   │   └── executions.js      # SSE status streaming
│   ├── utils/
│   │   ├── ai-fixer.js        # RCA & Selective Patching logic
│   │   ├── rag.js             # Local Embedding & Vector Search
│   │   ├── ingestor.js        # Background Ingestion Worker
│   │   ├── ast.js             # Tree-sitter Code Chunking
│   │   └── events.js          # Server-Sent Events (SSE) manager
│   └── server.js              # Entry Point
├── frontend/
│   └── src/
│       ├── pages/             # Dashboard, Pipelines (Real-time monitoring)
│       └── components/        # High-Fidelity UI components
└── README.md

---

made with ❤️ by Team CodeImpact