Noizefield/audio-plugin-coder

What it solves

Building professional audio plugins (VST3, AU, CLAP) is traditionally complex, requiring deep knowledge of Digital Signal Processing (DSP), UI frameworks, build systems, and packaging. Audio Plugin Coder (APC) lowers this barrier by providing a structured, AI-first framework that allows users to create plugins using natural language and guided workflows.

How it works

APC acts as an agent-agnostic system that provides LLM coding agents (like Claude Code, Cursor, or Codex) with the necessary domain knowledge and structure to build plugins. It uses a five-phase workflow:

1. Setup: Configures the toolchain and model preferences.
2. Dream: Ideation and creation of a creative brief.
3. Plan: Architecture design and UI framework selection.
4. Design: GUI mockup and visual iteration.
5. Implement: Coding the DSP engine and integrating the UI.
6. Ship: Packaging and distributing the final product.

The system tracks progress via a status.json file for each project, allowing users to resume work or rollback errors. It also features a self-improving troubleshooting system that captures and documents resolutions to new errors to improve future performance.

Who it’s for

It is designed for music producers, sound designers, and developers who want to build audio software instruments and effects without needing to be experts in the JUCE framework or complex C++ audio development.

Highlights

• Agent Agnostic: Works with various LLM agents including Claude Code, Kilo, and Cursor.
• Dual UI Options: Supports both native C++ (Visage) and web-based (HTML/CSS/JS via JUCE WebView) interfaces.
• Cross-Platform: Supports Windows, macOS, and Linux.
• Comprehensive Templates: Includes bridge templates for FFGL visual plugins and Max/MSP externals.
• Self-Learning: Automatically documents troubleshooting steps to build a local knowledge base of resolutions.