Multi-Agent Digital Audio

MAGDA is a free, open-source DAW with AI integrated from the ground up. Built on C++20, JUCE, and Tracktion Engine.

English | 简体中文

Features

Hybrid tracks: every track hosts both audio and MIDI clips
Three views: Arrangement, Session, and Mix
AI chat: natural language commands that generate and execute a custom DSL directly in the app (BYOAK, Bring Your Own API Keys)
Modulation system: 16 LFOs (with bezier curve editor) and 16 macro knobs per device and rack
Racks: parallel processing chains with volume, pan, mute, solo per chain, fully nestable
Piano roll with pitchbend and MIDI CC lanes
Drum grid device
Session view with clip launching
Mixer with faders, pan, mute, solo, sends, and I/O routing
Collapsible, resizable panels: Inspector + AI Chat (left), Plugin Browser + Sample Browser (right), context-sensitive editor (bottom)

Status

See Issues for known bugs and planned features.

Building

Prerequisites

C++20 compiler (GCC 10+, Clang 12+, or Xcode)
CMake 3.20+
Git LFS — required to fetch bundled binary assets
(CJK font, etc.). Install with brew install git-lfs (macOS),
apt install git-lfs (Debian/Ubuntu), or choco install git-lfs (Windows),
then run git lfs install once per machine.
Windows only: vcpkg supplies libxml2
(there is no system copy on Windows). Clone and bootstrap it once, then set a
VCPKG_ROOT environment variable pointing at the checkout — the build
auto-detects it and installs libxml2 from the vcpkg.json manifest on first
configure. On macOS/Linux libxml2 comes from the system, so no vcpkg is needed.
```
git clone https://github.com/microsoft/vcpkg C:\vcpkg
C:\vcpkg\bootstrap-vcpkg.bat
setx VCPKG_ROOT C:\vcpkg   # reopen the shell so it takes effect
```

Quick Start

# Clone with submodules and LFS assets
git clone --recursive https://github.com/Conceptual-Machines/magda-core.git
cd magda-core
git lfs pull  # safety net if git-lfs wasn't installed at clone time

# Setup and build
make setup
make debug

# Run
make run

Make Targets

make setup      # Initialize submodules and dependencies
make debug      # Debug build
make release    # Release build
make test       # Run tests
make clean      # Clean build artifacts
make format     # Format code
make lint       # Run clang-tidy analysis

Automated Workflows

The project includes automated GitHub Actions workflows:

CI Workflow: Runs on every push to validate builds and code quality
Security Scanning: CodeQL analysis, secret detection, and vulnerability scanning
Periodic Code Analysis: Weekly scans for TODOs, FIXMEs, and code smells
Refactoring Scanner: Bi-weekly analysis of code complexity and technical debt

See docs/ci/automated-workflows.md for details on automated analysis and periodic workflows.

Security

MAGDA takes security seriously. The repository implements comprehensive security measures:

🔒 Branch Protection: Main branch protected with required reviews and status checks
🔍 Automated Scanning: CodeQL security analysis for C++ vulnerabilities
🔐 Secret Detection: Automated scanning to prevent credential leaks
🛡️ Dependency Monitoring: Dependabot for security updates
⚡ CI/CD Security: All security checks must pass before merge

Found a security issue? Please review our Security Policy for responsible disclosure.

For detailed information about branch protection and security architecture, see docs/ci/branch-protection.md.

Architecture

magda/
├── daw/        # DAW application (C++/JUCE)
│   ├── audio/      # Audio processing
│   ├── core/       # Track, clip, selection management
│   ├── engine/     # Tracktion Engine wrapper
│   ├── interfaces/ # Abstract interfaces
│   ├── profiling/  # Performance profiling
│   ├── project/    # Project management and serialization
│   ├── ui/         # User interface components
│   └── utils/      # Utility helpers
└── agents/     # Agent system (C++)
tests/          # Test suite
scripts/        # Development and build scripts
docs/           # Documentation

Dependencies

Tracktion Engine - Audio engine
JUCE - C++ application framework (GUI, audio I/O, plugin hosting, MIDI, DSP)
juce-llm - LLM API client module
llama.cpp - Embedded local LLM inference
Catch2 - Testing (fetched via CMake)

Issues

Heads up: MAGDA is in early v0. Development started in January 2026 — internal iteration first, public release more recently — and is very much active. Expect bugs and missing pieces. The best way to help the project is to file an issue.

Found a bug or have a feature request? Please open an issue on GitHub.

License

GPL v3 - see LICENSE for details.

magda-core