The Muser – Open-source alternative to Suno, runs locally, you own everything

The Muser, an open-source music generation tool, now allows users to compose music locally on their own hardware using natural language descriptions, eliminating the need for subscriptions or API keys. The system orchestrates multiple AI models—including NotaGen for notation, ACE-Step for audio, and DiffSinger for vocals—to produce commercially-safe, production-ready music with full user ownership. By running entirely on local hardware with an NVIDIA GPU or CPU, The Muser provides a free alternative to services like Suno and ElevenLabs Music, with all output licensed for commercial use under Apache 2.0 or MIT terms.

The open-source alternative to Suno and ElevenLabs Music. Run locally. Own everything. No subscriptions, no ToS, no limits. Describe what you want to hear in natural language, and The Muser orchestrates AI models to produce scores, audio, and vocal performances — entirely on your hardware. pip install -e "." ollama pull qwen3:30b-a3b bash scripts/setup models.sh muser You:Compose a 2-minute lo-fi hip hop beat with jazzy piano and vinyl crackleThe Muser generates candidates, selects the best, and exports production-ready audio. User ──► LLM Agent 46 tools ──► AI Models ──► Your Music │ │ Plans, validates, NotaGen notation iterates, mixes ACE-Step audio DiffSinger vocals RVC voice cloning Natural language composition — describe music in plain English, get professional output 46-tool vocabulary — generation, validation, rendering, voice, effects, mixing, curation Multiple AI models — NotaGen for classical notation, ACE-Step for modern audio, DiffSinger for singing Full voice pipeline — RVC voice conversion, Demucs stem separation, feminization presets Quality scoring — 9-metric analysis with letter grades, best-of-N candidate selection Audio-to-MIDI bridge — extract sheet music from generated audio Individual effects — EQ, reverb, compression, volume, mixing — all controllable by the LLM 12-dimension curation — hard gates + soft scores for batch quality control Web UI — Gradio interface with chat, audio player, and composition status Streaming — token-by-token LLM responses, no more staring at spinners Every default generation path produces commercially-safe output: | Path | License | Commercial Use | |---|---|---| | ACE-Step audio | Apache 2.0 | YES | | NotaGen notation | MIT | YES | | DiffSinger + Griffin-Lim default | Apache 2.0 | YES | | RVC voice conversion | MIT | YES | | FluidSynth/sfizz rendering | LGPL/BSD | YES | See docs/legal.md /noah-chelednik/the-muser/blob/main/docs/legal.md for the full breakdown including optional components. - Python 3.10+ - NVIDIA GPU with 24GB VRAM for full pipeline or CPU-only LLM orchestration Ollama https://ollama.com for local LLM inference- ffmpeg, FluidSynth for audio rendering git clone https://github.com/noah-chelednik/the-muser.git cd the-muser python -m venv .venv && source .venv/bin/activate pip install -e ". gpu,voice " GPU + voice pipeline bash scripts/setup environment.sh System tools + Ollama bash scripts/setup models.sh AI model weights cp .env.example .env Edit as needed docker-compose up muser-gpu GPU mode requires NVIDIA Container Toolkit docker-compose up muser-cpu CPU-only mode docker-compose up muser-web Web UI at http://localhost:7860 muser Interactive session muser -c my-piece Resume a composition muser --stream Streaming responses default muser -m groq/llama-3.3-70b Use a specific LLM provider muser-web Launch at http://localhost:7860 No paid API key required. The Muser routes to the best available provider: | Provider | Speed | Cost | Setup | |---|---|---|---| | Groq | 300+ tok/s | Free tier | Set GROQ API KEY | | Cerebras | 1000+ tok/s | Free tier | Set CEREBRAS API KEY | | Gemini | Fast | Free tier | Set GOOGLE API KEY | | Ollama local | 12-28 tok/s | Free forever | ollama pull qwen3:30b-a3b | src/ orchestrator/ LLM agent loop, 46 tools, composition state generation/ AI model wrappers NotaGen, ACE-Step, DiffSinger audio/ Rendering, validation, effects, mixing, export notation/ Format conversion, theory validation, score rendering voice/ Voice conversion RVC, Seed-VC , stem separation curation/ 12-dimension quality analysis, batch curation web/ Gradio web interface See docs/architecture.md /noah-chelednik/the-muser/blob/main/docs/architecture.md for the full system design. pytest tests/ -v -m "not gpu" 258+ tests, ~6 seconds See CONTRIBUTING.md /noah-chelednik/the-muser/blob/main/CONTRIBUTING.md for development setup, code style, and the tool-adding guide. MIT /noah-chelednik/the-muser/blob/main/LICENSE — The Muser framework and all original code. See THIRD PARTY LICENSES.md /noah-chelednik/the-muser/blob/main/THIRD PARTY LICENSES.md for component licenses and docs/legal.md /noah-chelednik/the-muser/blob/main/docs/legal.md for output ownership details.