Automation system for Suno music creation, mastering, and export using a hybrid architecture:
- deterministic browser skills for reliable execution
- planner-driven orchestration for autonomous runs
- optional AI-assisted spec generation
We are actively building and hardening the in-repo hybrid autopilot:
- input: high-level
music_type - auto-generate: lyrics, styles, weirdness, style influence
- execute: create -> wait -> master -> export
- resilience: retries, checkpoint/resume, phase artifacts
Project defaults are configured for local Ollama:
- provider:
ollama - default model:
qwen3:8b - fast fallback option:
qwen3:0.6b
Example:
cd suno_mastering_agent
source venv/bin/activate
python main.py autopilot --music-type "edm" --count 3Phase 2 planning mode (parallel candidates + artifacts):
python main.py autopilot \
--music-type "edm" \
--count 3 \
--phase2 \
--candidate-count 4 \
--phase2-artifact-log /tmp/suno_phase2_artifacts.jsonlResumable robust mode:
python main.py autopilot \
--music-type "edm" \
--count 20 \
--step-retries 3 \
--checkpoint-file /tmp/suno_autopilot_checkpoint.json \
--resume \
--continue-on-errorThese are tracked as future enhancements, not required for the current build:
- LangChain DeepAgents integration (optional planner backend migration once executor reliability is stable)
- BMAD Method integration (process/workflow templates for planning discipline)
- Gastown-style orchestration (larger multi-agent workspace lifecycle management)
Decision: we will keep using lightweight in-repo equivalents first, then consider full external adoption once execution reliability is consistently strong.
suno_mastering_agent/main.py- CLI entry point (autopilot,autocreate,agent, etc.)suno_mastering_agent/src/skills/- deterministic browser action primitivessuno_mastering_agent/src/agents/- composed automation runnerssuno_mastering_agent/src/agent/- LLM/tool workflowssuno_mastering_agent/src/ui/- Gradio UI