AutoDev Agent — Autonomous AI Coder for VS Code

Intent → Plan → Execute → Done. No manual coding for boilerplate, debugging, or scaffolding.

---

What It Does

AutoDev Agent is a multi-agent AI system built inside VS Code. It replaces 2+ hours of developer work with a single command — no monthly subscription, no cloud lock-in, fully open source.

Feature	What happens
Fix Entire Project	Scans all files, detects every bug, fixes everything, shows diff before applying
Generate from Intent	Type "Build login system with JWT" → files created, code written, ready to run
Fix & Explain Selection	Select broken code → get explanation + working fix side by side
Refactor File	Choose goals (performance, readability, security) → production-ready code
Generate Tests	Full test suite written automatically for any file

---

Quick Start

1. Install the extension

From VS Code Marketplace (once published):

Ctrl+P → ext install autodev.autonomous-dev-agent

From source:

git clone https://github.com/yourusername/autonomous-dev-agent
cd autonomous-dev-agent
npm install
npm run compile
# Press F5 in VS Code to launch Extension Development Host

2. Configure your LLM provider

Open the Command Palette (Ctrl+Shift+P) and run:

AutoDev: Configure LLM Provider

Pick one of three free options:

---

Free LLM Options

Option A: Ollama (Recommended — 100% local, 100% free)

# 1. Install Ollama
curl -fsSL https://ollama.com/install.sh | sh   # Linux/Mac
# Windows: https://ollama.com/download

# 2. Pull a coding model (choose one)
ollama pull codellama:7b          # Best for general coding
ollama pull deepseek-coder:6.7b   # Best for code generation
ollama pull mistral:7b            # Best for explaining/refactoring
ollama pull qwen2.5-coder:7b      # Excellent all-rounder (recommended)

# 3. Start Ollama
ollama serve

Then in VS Code settings set:

"autodev.provider": "ollama",
"autodev.ollamaModel": "qwen2.5-coder:7b"

Option B: Google Gemini (Free tier — 1500 requests/day)

1. Go to aistudio.google.com
2. Click Get API Key → Create key (free)
3. Run AutoDev: Configure LLM Provider → Select Gemini → Paste key

"autodev.provider": "gemini",
"autodev.geminiApiKey": "AIza...",
"autodev.geminiModel": "gemini-1.5-flash"

Option C: LM Studio / LocalAI (OpenAI-compatible)

# Download LM Studio: https://lmstudio.ai
# Load any GGUF model, start local server on port 1234

"autodev.provider": "openai-compatible",
"autodev.openaiCompatibleUrl": "http://localhost:1234/v1",
"autodev.openaiCompatibleModel": "local-model"

---

Keyboard Shortcuts

Shortcut	Action
Ctrl+Shift+F / Cmd+Shift+F	🔥 Fix Entire Project
Ctrl+Shift+G / Cmd+Shift+G	⚡ Generate from Intent
Ctrl+Shift+A / Cmd+Shift+A	🤖 Open Agent Panel

---

Usage Examples

Generate from Intent

Open Command Palette → AutoDev: Generate from Intent

"Build a REST API with Express, JWT auth, and PostgreSQL"
"Create a React dashboard with charts and dark mode"
"Make a Python CLI tool that converts CSV to JSON"
"Build a Discord bot that answers questions with AI"

The Planner Agent shows you exactly what it will create before executing.

Fix Entire Project

Ctrl+Shift+F  →  Confirm  →  Watch it scan every file  →  Review diff  →  Apply All

Fix & Explain Selection

1. Select any code (or nothing for the whole file)
2. Right-click → AutoDev: Fix & Explain Selection
3. See: explanation, original, fixed code — in tabs
4. Click Apply Fix

Refactor Current File

1. Open a file
2. Command Palette → AutoDev: Refactor Current File
3. Pick goals: Performance, Readability, Security, Clean code, Modern syntax
4. Review changes → Apply

---

Architecture

User Input
    │
    ▼
┌─────────────────────────────────────────────────┐
│  VS Code Extension Layer                        │
│  Commands + Sidebar Panel + Status Bar          │
└─────────────────────────────────────────────────┘
    │
    ▼
┌─────────────────────────────────────────────────┐
│  Multi-Agent System                             │
│                                                 │
│   Planner Agent                               │
│     Intent → Structured plan with steps        │
│                                                 │
│   Executor Agent                              │
│     Creates files, runs commands, installs pkgs │
│                                                 │
│   Debugger Agent                              │
│     Finds bugs, fixes code, writes tests        │
│                                                 │
│   Validator Agent                             │
│     Checks quality (0-100 score) before apply  │
└─────────────────────────────────────────────────┘
    │
    ▼
┌─────────────────────────────────────────────────┐
│  LLM Provider Layer                             │
│                                                 │
│  Ollama ──── local, free, private               │
│  Gemini ──── free tier (1500 req/day)           │
│  OpenAI-compatible ─── LM Studio, Groq, etc.   │
└─────────────────────────────────────────────────┘

---

Full Configuration Reference

{
  // Provider: "ollama" | "gemini" | "openai-compatible"
  "autodev.provider": "ollama",

  // Ollama settings
  "autodev.ollamaUrl": "http://localhost:11434",
  "autodev.ollamaModel": "qwen2.5-coder:7b",

  // Gemini settings
  "autodev.geminiApiKey": "",
  "autodev.geminiModel": "gemini-1.5-flash",

  // OpenAI-compatible settings
  "autodev.openaiCompatibleUrl": "http://localhost:1234/v1",
  "autodev.openaiCompatibleKey": "local",
  "autodev.openaiCompatibleModel": "local-model",

  // Behaviour
  "autodev.maxFilesPerScan": 50,
  "autodev.showDiffBeforeApply": true,
  "autodev.autoRunTests": false,
  "autodev.excludePatterns": [
    "node_modules", ".git", "dist", "build", "out", ".next", "coverage"
  ]
}

---

Development

# Clone
git clone https://github.com/yourusername/autonomous-dev-agent
cd autonomous-dev-agent

# Install dependencies
npm install

# Compile TypeScript
npm run compile

# Watch mode (auto-recompile on save)
npm run watch

# Open in VS Code and press F5 to launch Extension Development Host
code .

Build a .vsix package

npm install -g @vscode/vsce
vsce package
# → autonomous-dev-agent-1.0.0.vsix

Install the .vsix locally

VS Code → Extensions → ⋯ → Install from VSIX

---

Recommended Ollama Models

Model	Size	Best for
qwen2.5-coder:7b	4.7GB	All-round code generation ⭐
deepseek-coder:6.7b	3.8GB	Fast code completion
codellama:7b	3.8GB	Python / general coding
mistral:7b	4.1GB	Explanation & refactoring
phi3:mini	2.2GB	Low RAM machines
llama3.2:3b	2.0GB	Very fast, basic tasks

---

Privacy

• Ollama mode: All code stays on your machine. Zero data leaves your system.
• Gemini mode: Code is sent to Google's API. Review Google's privacy policy.
• The extension never stores your code or API keys in any external service.

---

Contributing

PRs welcome! Key areas to improve:

• More agent types — Architect Agent, Documentation Agent, Migration Agent
• Self-healing loop — wire terminal error output back into Debugger Agent
• Better diff UI — inline diff with VS Code's native diff viewer
• Model benchmarks — test which models perform best per task type
• Language-specific prompts — tuned system prompts per language

See CONTRIBUTING.md to get started.

---

License

MIT — free to use, modify, distribute, and sell.

---

Star History

If AutoDev saves you time, a GitHub star helps others find it!

---

Built with ❤️ for developers who want to feel like they're coding in 2035.