Ducky - Mentor de Código com RAG Local
FelixSoderstrom/Ducky
Monitor inteligente de código que ensina boas práticas sem escrever código
Ver no GitHubMCP server que monitora mudanças no seu codebase em tempo real e fornece feedback educativo sobre qualidade de código, padrões e boas práticas. Funciona como um mentor ao ombro, analisando via RAG sem gerar código, apenas ensinando por que algo é problemático. Totalmente local com TTS nativo.
Instalação
npx convez-ai FelixSoderstrom/Ducky1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Code review automatizado para desenvolvedores solo
- Ferramenta de aprendizado para desenvolvedores juniores
- Identificação de débito técnico em tempo real
- Padronização de código em grandes bases
Veredito

Ducky resolve um problema real: ferramentas de IA geram código rapidamente mas deixam devs sem entender. Seu foco em educação via RAG + feedback contextual é diferenciado e genuinamente útil para devs que querem evoluir, não apenas completar tarefas. Infraestrutura local-first com TTS elimina dependências externas.
README
A tradução para o português deste README ainda está na fila.
Ducky - Local Code Mentor with RAG-Powered Feedback
A file-watching assistant that analyzes your codebase changes and provides real-time feedback on code quality, patterns, and best practices.
Quick Start
pip install -r requirements.txt
python main.py
What It Does
- Monitors your codebase for changes via file watchers
- Maintains a local SQLite database of your code structure
- Uses RAG (Retrieval-Augmented Generation) to analyze code changes
- Provides contextual feedback on code quality and patterns
- Acts as a code review mentor, not an autocomplete tool
Key Features
- Local-first: All data stays on your machine
- Real-time analysis: Immediate feedback on code changes
- Context-aware: Understands your entire codebase structure
- Educational focus: Explains why something is problematic
- Non-intrusive: Suggests improvements without writing code for you
- Local TTS: Uses Chatterbox TTS for voice notifications
Architecture Overview
File System Changes → File Watcher → SQLite Database → RAG Analysis → LLM Feedback
Use Cases
- Code review automation for solo developers
- Learning tool for junior developers
- Consistency enforcement across large codebases
- Technical debt identification in real-time
Technical Stack
- File system monitoring
- SQLite for local code indexing
- RAG pipeline for contextual analysis
- LLM integration for feedback generation
- Chatterbox TTS for local voice synthesis
Detailed Documentation
Problem Statement
Modern AI coding tools excel at code generation but often lead to poor practices when developers don't understand the generated code. Ducky bridges this gap by focusing on education and code quality rather than code generation.
How It Works
- File Monitoring: Watches your project directory for changes
- Code Indexing: Maintains a searchable database of your codebase
- Change Analysis: Uses RAG to understand the context of your changes
- Feedback Generation: LLM analyzes changes against best practices
- Developer Notification: Provides actionable feedback and explanations
Voice Notifications
Ducky uses Chatterbox TTS for voice notifications:
- No API costs - Runs completely locally
- No API keys required - Zero external dependencies
- Voice cloning - Support for custom voice prompts
- Emotion control - Adjustable exaggeration and intensity
I never recommend cloning the dev branch and will never promise working versions on the dev branch. To get a sneak peek of what coming around the corner, check out my Miro Board
Philosophy
Ducky doesn't write code for you - it teaches you to write better code. The goal is to maintain developer agency while providing expert-level guidance on code quality and patterns.