Spitcode

From spoken idea to production-ready FastAPI app — powered by local LLMs.

Spitcode is a full pipeline that transforms a simple spoken idea into production-hardened FastAPI code with a complete README.md, error handling, rate limiting, logging, and OpenAPI docs. No cloud. No API keys. All offline.

🚀 Quick Start

# Install dependencies
pip install -r requirements.txt

# Start Ollama with Qwen model
ollama run qwen3:14b-q4_K_M

# Build your first app
python -m spitcode.main build

✨ Features

• 🎤 Voice Input: Record your ideas naturally
• 🤖 Local LLM: Powered by Qwen via Ollama
• 🔒 Security First: Built-in rate limiting, error handling, and security best practices
• 📝 Auto-Documentation: Generates comprehensive README with curl examples
• 🔄 Code Improvement: Automatic code analysis and refactoring
• 🛠️ Production Ready: Includes middleware, logging, and error handling
• 📚 OpenAPI Docs: Automatic API documentation generation

🎯 Why Spitcode?

Because ideas are cheap, but building them isn't. Spitcode helps solo devs and startup builders go from "I want to build an app that does X" to working backend code in minutes.

And it all runs on your machine. With your models. Your voice. Your ideas.

🧠 What It Does

Spitcode is a modular AI pipeline built for:

• Recording a spoken feature idea
• Generating initial FastAPI code
• Analyzing and improving structure, security, and style
• Hardening the code for production (middleware, secrets, error handling, etc.)
• Auto-generating a README.md with curl examples and setup

🛠️ Requirements

• Python 3.10+
• Ollama with Qwen model
• FFmpeg (for audio processing)
• 8GB+ RAM recommended
• 20GB+ free disk space

⚡ Pipeline Overview

Step	Command	Description
1.1	python -m spitcode.main record "a todo app with auth and billing"	Records voice and saves transcript.
1.2	python -m spitcode.main generate	Generates first main.py.
2.1	python -m analyzer.code_analyzer	Analyzes code and proposes improvements.
2.2	python -m analyzer.parse_analysis	Converts feedback into structured JSON.
3.1	python improver/code_improver.py	Applies structured improvements.
3.2	python improver/hardener_agent.py	Adds prod-grade middleware, error handling, logging.
3.3	python improver/generate_readme.py	Generates install + usage docs from code.
Final	python -m spitcode.main build	Runs all the above in one go.

🧪 Example Use Case

You say:

"Build a todo app with login, JWT, SQLite, and subscription billing."

Spitcode will generate:

• main.py — working FastAPI app
• main_rewritten.py — refactored version with improvements
• main_hardened.py — final production-ready API
• README.md — curl examples, installation, API docs

🧬 Architecture

• Local LLM: Uses Qwen via Ollama
• Voice Input: Whisper + FFmpeg
• Analysis: RAG with your local analyzer
• Improvement: Prompt-based refactor with safety guards
• Harden: Security best practices (envvars, headers, rate limiting)
• Documentation: Auto-generated README with curl samples

📁 Output Folder Structure

outputs/
├── transcript.txt           # What you said
├── main.py                 # First LLM-generated code
├── main_rewritten.py       # Improved/refactored
├── main_hardened.py        # Prod-ready code
├── review_chunks.json      # Improvement suggestions
└── README.md               # Auto-written documentation

🔧 Troubleshooting

Common Issues

1. Ollama not starting

   • Ensure Ollama is installed and running
   • Check if the Qwen model is downloaded
   • Verify system requirements are met

2. Audio recording fails

   • Check FFmpeg installation
   • Verify microphone permissions
   • Ensure audio input device is working

3. Code generation issues

   • Clear the outputs directory
   • Check Python version (3.10+ required)
   • Verify all dependencies are installed

🗺️ Roadmap

• Support for multiple LLM backends
• Web UI for easier interaction
• Custom template support
• Database migration generation
• Test case generation
• Docker support
• CI/CD pipeline generation