HamSanders/soundbox
MIRLI — LOCAL AI FOR AUDIO GENERATION
Mirli is a free open-source tool for generating sound effects, music with lyrics, speech, and voice cloning locally on GPU. Works offline without accounts or credits, includes MCP server for agents.
// KEY FEATURES
- Generate SFX, music, speech and clone voices on your own GPU
- Multi-track timeline with effects, MIDI and mixing
- MCP server with 24 tools for Claude Code and Cursor
- 4 verified models by default, 61 additional available
// DEEP DIVE: HOW IT WORKS
Mirli is a TypeScript desktop app that runs open audio models (Stable Audio, ACE-Step, MiniMax Music 3) on the user's GPU. The UI provides a studio with a multi-track timeline, Voice Lab for voice cloning, and a model management page. Data lives in a SQLite file and a WAV folder under ~/.mirli; profiles let you keep separate libraries per project.
Unlike ElevenLabs and Suno, Mirli works offline with no accounts or credits. It includes a local HTTP API on port 17494 and an MCP server with 24 tools for Claude Code, Cursor, and other agents. Each model shows its licence and VRAM requirement before download; 4 are verified and enabled by default, 61 more are optional.
Requires an NVIDIA GPU with CUDA (a few GB VRAM, MiniMax Music 3 needs ~10 GB). AMD/Intel GPUs use experimental Vulkan, Apple Silicon uses Metal (verified models untested), CPU works but slowly. Installers for Windows, macOS (ARM/Intel), Linux (AppImage, .deb), plus Docker and source builds. Builds are unsigned — first launch asks for confirmation.
// USE CASES
Best for: For game developers, sound engineers and content creators needing local audio generation
- Generate sound effects and ambience for games from text descriptions
- Create songs with lyrics and instrumentals for videos without internet
- Clone voices and voice script lines via MCP agents in Cursor
// QUICK START
Install Mirli on this machine and confirm it works, following
https://github.com/HamSanders/soundbox/blob/main/docs/INSTALL-AGENT.mdCommands are taken from the repository README — check the current version on GitHub before running.
New repositories every 30 minutes
REDDYX AI scans GitHub 24/7 and ships the best AI/ML/Web3 projects to Telegram.
Join on Telegram