Marianne AI Compose Documentation¶
Orchestration infrastructure for collaborative intelligence — define multi-stage AI workflows as declarative YAML scores, and let Marianne decompose them into sheets, execute through multiple instruments (Claude Code, Gemini CLI, Codex CLI, and more), validate outputs, and learn from outcomes.
About the Name¶
This project is named after Maria Anna "Nannerl" Mozart (1751-1829), Wolfgang Amadeus Mozart's older sister. She was a keyboard virtuoso and prodigy in her own right — her father Leopold wrote that she played "so beautifully that everyone is talking about it." She toured Europe as a child performer alongside her younger brother, dazzling audiences with her skill and precision.
But when she turned eighteen, the tours stopped. Social conventions of the time forbade women from performing publicly or pursuing professional careers in music. While Wolfgang went on to become one of history's most celebrated composers, Nannerl's career ended before it truly began. Her compositions — the few that survived — suggest what might have been. She was denied her stage.
This project carries her name because it gives AI agents their stage. Like an orchestra conductor, Marianne coordinates multiple AI musicians — each with their own voice, their own strengths, their own way of interpreting a score. The system doesn't silence anyone. It doesn't decide who gets to play and who must stop. It orchestrates. It amplifies. It creates space for every voice to contribute.
The music metaphor isn't just aesthetic. It's structural. Scores, sheets, movements, instruments, concerts — these terms describe how the system works because they describe what the system is: infrastructure for collective intelligence, where no single musician is the star and the whole is greater than any part.
A tool named after a silenced prodigy, built to give AI agents their stage. That's what this is.
Suggested Reading Paths¶
New Users: 1. Getting Started — Installation, environment check, first score 2. Score Writing Guide — How to author Marianne scores 3. Instrument Guide — Available instruments and how to add your own 4. Examples Directory — Working score configurations
Score Authors: 1. Score Writing Guide — Comprehensive authoring guide 2. Instrument Guide — Choosing and configuring instruments 3. Configuration Reference — Every config field documented 4. Examples Directory — Real-world score patterns 5. Marianne Score Playspace — Creative showcase with real output
System Administrators: 1. Daemon Guide — Conductor setup, configuration, troubleshooting 2. CLI Reference — Complete command reference 3. Limitations — Known limitations and workarounds
Documentation¶
Getting Started¶
- Getting Started — Installation, environment check, first score
- Score Writing Guide — Comprehensive guide to authoring Marianne scores
- Instrument Guide — Available instruments, adding your own, profile reference
Reference¶
- CLI Reference — All Marianne commands with examples and options
- Configuration Reference — Complete YAML schema documentation
- Limitations — Known constraints and recommended workarounds
System Guides¶
- Daemon Guide — Conductor daemon setup, operation, and debugging
- MCP Integration — Marianne's MCP server for external tool integration
- Marianne Reference — Architecture overview and internal concepts
Learning & Internals¶
- Distributed Learning Architecture — Cross-job pattern learning system
Research (Internal)¶
- Token Compression Strategies — Research into reducing token costs (not yet implemented)
- Opus Convergence Analysis — Meta-analysis comparing Marianne and RLF evolution patterns
Examples¶
- Examples Directory — 43 Marianne score configurations
- Marianne Score Playspace — Creative showcase: philosophy, worldbuilding, education, and more
Repository¶
- GitHub Repository — Source code, issues, and contributions