AI-powered framework for vibe-coding audio plugins from concept to shipped product
Audio Plugin Coder (APC) is the result of a long-standing personal obsession: building creative tools, writing music, and ultimately creating professional audio plugins.
While developing software instruments and effects has always been a dream, building real-world VSTs (with robust DSP, UI, state handling, and packaging) is notoriously complex. Over time, and especially with the rapid advancement of AI-assisted development, that barrier has finally crumbled.
Over the past 18 months, APC has been continuously designed, tested, and re-iterated as a practical AI-first framework for building audio plugins. This involved thousands of hours of experimentation, trial-and-error, and yes... occasionally yelling at LLMs to finally render the UI correctly.
Midway through development, I stumbled upon the excellent work of TÂCHES (glittercowboy). His approach to context engineering was a revelation. I adopted some of his core ideas, particularly regarding meta prompting and structured agent workflows and integrated them directly into APC's DNA to create a more robust system.
APC is designed to be Agent Agnostic. Whether you use Google's Antigravity, Kilo, Claude Code, or Cursor, APC provides the structure they need to succeed.
Audio Plugin Coder (APC) is currently in active development. APIs may change, features may be incomplete, and bugs should be expected.
Use APC for development and experimentation purposes only until a stable release is announced.
Audio Plugin Coder (APC) is a structured, AI-driven workflow system that guides LLM agents through the entire audio plugin development lifecycle.
It enables the creation of VST3 / AU / CLAP plugins using natural language, predefined workflows, and domain-specific skills- without constantly re-explaining context, architecture, or best practices to the AI.
Instead of manually juggling DSP architecture, UI frameworks, build systems, state tracking, and packaging, APC provides a unified framework where AI agents can operate with long-term context, validation, and self-improving knowledge.
- 🤖 LLM-Driven Development - Designed to work with Antigravity, Kilo, Claude Code, Cursor, or any coding agent.
- 🎯 Structured Workflows - Five-phase system: Dream → Plan → Design → Implement → Ship.
- 🎨 Dual UI Frameworks - Choose Visage (pure C++) or WebView (HTML5 Canvas).
- 📊 State Management - Automatic progress tracking, validation, and rollback capabilities.
- 🔧 Self-Improving - Auto-capture troubleshooting knowledge; the system gets smarter over time.
- 🏗️ Production Ready - JUCE 8 integration with CMake build system.
- 📚 Comprehensive Skills - Pre-built domain knowledge for DSP, UI design, testing, and packaging.
- Windows 11 (macOS/Linux via GitHub Actions)
- PowerShell 7+
- Visual Studio 2022 (with C++ development tools)
- CMake 3.22+
- Git
- An LLM coding agent (Claude Code, Kilo, etc.)
- Clone the repository (with submodules):
git clone --recursive https://github.com/Noizefield/audio-plugin-coder.git
cd audio-plugin-coderOr clone normally and run setup:
git clone https://github.com/Noizefield/audio-plugin-coder.git
cd audio-plugin-coder
.\scripts\setup.ps1- Initialize your LLM agent:
For Kilo:
# Workflows are automatically discovered from .kilocode/workflows/For Claude Code:
# The agent will discover workflows from .kilocode/workflows/- Create your first plugin:
/dream MyReverb
The AI will guide you through the entire process!
💭 DREAM (Ideation)
↓ Create creative brief, define parameters
📋 PLAN (Architecture)
↓ Design DSP graph, select UI framework
🎨 DESIGN (GUI)
↓ Create mockups, iterate on visual design
💻 IMPLEMENT (Code)
↓ Build DSP engine, integrate UI
🚀 SHIP (Package)
↓ Build installers, test in DAWs
| Command | Description |
|---|---|
/dream [Name] |
Start new plugin with ideation phase |
/plan [Name] |
Define architecture and select UI framework |
/design [Name] |
Create GUI mockups and visual design |
/impl [Name] |
Implement DSP and UI code |
/ship [Name] |
Package and distribute plugin |
/status [Name] |
Check current progress and state |
/resume [Name] |
Continue development from last phase |
/new [Name] |
Run complete workflow with confirmations |
You: /dream TapeDelay
AI: Great! Let's define your plugin concept.
What is the sonic character?
(e.g., "Warm analog tape delay with wow/flutter")
You: Vintage tape echo with analog saturation
AI: Perfect! What are the top 3-5 parameters you want?
You: Delay time, feedback, wow/flutter amount, saturation, mix
AI: ✅ Dream phase complete!
Files created:
- plugins/TapeDelay/.ideas/creative-brief.md
- plugins/TapeDelay/.ideas/parameter-spec.md
- plugins/TapeDelay/status.json
Next step: /plan TapeDelay
APC uses a unique State Management system (status.json) to track development across phases. This allows you to pause, switch agents, or restart specific phases without losing context.
audio-plugin-coder/
├── .[Agent]/ # AI agent configuration
│ ├── workflows/ # Slash command orchestrators
│ │ ├── dream.md
│ │ ├── plan.md
│ │ ├── design.md
│ │ ├── impl.md
│ │ └── ship.md
│ ├── skills/ # Domain knowledge modules
│ │ ├── skill_ideation/
│ │ ├── skill_planning/
│ │ ├── skill_design/
│ │ ├── skill_implementation/
│ │ └── skill_packaging/
│ ├── guides/ # Reference documentation
│ │ └── state-management-guide.md
│ ├── rules/ # System constraints
│ │ ├── agent.md
│ │ └── file-naming-conventions.md
│ ├── troubleshooting/ # Auto-captured issues
│ │ ├── known-issues.yaml
│ │ └── resolutions/
│ └── templates/
├── docs/ # Comprehensive documentation
├── plugins/ # Generated plugins
│ └── [YourPlugin]/
│ ├── .ideas/ # Specs and planning
│ ├── Design/ # UI mockups
│ ├── Source/ # C++ code
│ └── status.json # State tracking
├── scripts/ # Build automation
│ ├── state-management.ps1
│ └── build-and-install.ps1
└── build/ # Compilation artifacts
Skills contain domain knowledge (the "how"):
- Step-by-step instructions
- Best practices
- Framework-specific guidance
- Code generation patterns
Workflows orchestrate skills (the "when"):
- Prerequisites validation
- Phase transitions
- State management
- Error recovery
Example: The /design workflow checks your UI framework selection (Visage or WebView) from status.json, then loads the appropriate design skill automatically.
- Native JUCE components
- High performance
- Full C++ control
- Custom rendering with
visage::Frame
Note: Visage integration is planned but not yet implemented in the current beta version.
- Modern web technologies
- Rapid iteration with hot reload
- Rich component libraries
- Canvas-based rendering for performance
The AI helps you choose based on your plugin's complexity and requirements during the planning phase.
Every plugin has a status.json file tracking:
- Current development phase
- UI framework selection
- Completed milestones
- Validation checkpoints
- Error recovery points
Benefits:
- Resume development any time
- Validate prerequisites automatically
- Rollback on errors
- Track project history
APC includes an auto-capture system that learns from problems:
- AI encounters error → Searches known issues database
- If known → Applies documented solution immediately
- If unknown → Attempts resolution, tracks attempts
- After 3 attempts → Auto-creates issue entry
- When solved → Documents solution for future use
Location: .kilocode/troubleshooting/
Result: The system gets smarter with every issue encountered!
APC works with any LLM-based coding agent that supports:
- Custom workflows/slash commands
- File system access
- PowerShell execution
Tested with:
- ✅ Claude Code (Anthropic)
- ✅ Kilo (kilo.ai)
- Cursor
- Others welcome!
- JUCE 8 - Audio plugin framework
- CMake - Build system
- PowerShell - Automation scripting
- JUCE 8 - Audio plugin framework (includes DSP, GUI, etc.)
- WebView2 - Chromium-based web UI
- YAML - Knowledge base format
- Markdown - Documentation and workflows
| Format | Windows | macOS | Linux |
|---|---|---|---|
| VST3 | ✅ | ✅ | ✅ |
| Standalone | ✅ | ✅ | ✅ |
| AU | ❌ | ✅ | ❌ |
| LV2 | ❌ | ❌ | ✅ |
CLAP support planned for future release.
Comprehensive documentation is available in the docs/ directory:
- Getting Started - Documentation index and quick start
- Plugin Development Lifecycle - Detailed phase guide
- Command Reference - All commands and scripts
- FAQ - Frequently asked questions
- Troubleshooting - Common issues and solutions
- Windows support
- GitHub Actions CI/CD
- Comprehensive documentation
- macOS local build support
- Linux local build support
- CLAP format support
- Preset management system
- Plugin marketplace integration
- Real-time collaboration features
I am an independent developer pouring hundreds of hours (and significant API costs) into this project.
Developing a framework that works across different AI agents means constantly testing against paid tiers of Claude, Gemini, and others. I often run out of "Plan" usage just testing a single workflow improvement.
If APC saves you time, helps you learn JUCE, or helps you ship a plugin, please consider supporting the development. It helps cover API costs and accelerates macOS support!
☕ Buy Me a Coffee / Sponsor on GitHub
Crypto/Other options TBD
Contributions are welcome! Join our GitHub Discussions to connect with the community.
- Add Skills: Create new domain knowledge modules
- Test Platforms: Verify compatibility with different AI agents
- Improve Docs: Help us improve documentation
- Share Plugins: Showcase what you've built
See CONTRIBUTING.md for details.
- JUCE Team - For the industry-standard framework.
- The AI Community - Specifically the meta-prompting pioneers.
- TÂCHES (glittercowboy) - Inspiration for context engineering systems.
This project is licensed under the MIT License - see the LICENSE file for details.
Built with ❤️ (and a lot of tokens) for the audio development community.
Turn your plugin ideas into reality with the power of AI