Soundbox

Soundbox

The free, open-source, local alternative to ElevenLabs and Suno.

Sound effects, music, songs with lyrics, speech and voice cloning from open-weight models on your own GPU. Arrange it on a timeline, or let Claude Code, Cursor or any MCP agent do it. No account, no credits, works offline. MIT.

Download · Hear it · For agents · Models · Docs · Website

Watch: 30-second tour · Under the hood · Agents and MCP

Describe it, hear it, arrange it

Four models are verified end to end and on by default; 61 more can be switched on from the Models page. Each model's licence is shown before you download it. Everything stays in one SQLite file and a folder of WAVs under ~/.soundbox. Profiles give you several libraries in one install, one per game or client, switched from the header without a restart; the models are shared.

Download

The installer carries its own runtime. The builds are not signed yet, so the first launch asks you to confirm (where). Open SFX, describe the material, the action and the space, and generate: the first model downloads when you pick it, with its licence shown first.

Linux, Docker and running from source are under Run from source.

Install with a prompt

Paste this into Claude Code, Cursor or Codex:

Install Soundbox on this machine and confirm it works, following
https://github.com/HamSanders/soundbox/blob/main/docs/INSTALL-AGENT.md

The guide tells the agent to check the GPU and VRAM first, to show you each model's size and licence before downloading it, never to accept a non-commercial licence for you, and to finish by generating one sound and telling you where the WAV is.

Hear it

Each clip was generated by Soundbox from the prompt shown. The samples site has around 600 more.

For agents

Start Soundbox, then add it to your client.

claude mcp add --transport http soundbox http://127.0.0.1:17494/mcp

Cursor and other MCP clients take { "type": "http", "url": "http://127.0.0.1:17494/mcp" }. Then ask: "add a door slam and a coin pickup to this game", "voice these three lines as two characters", "write a 20-second loading-screen loop". The agent gets 24 tools for sounds, speech and multi-track stories, and three skills: soundbox-sfx, soundbox-sound-pass and soundbox-voice-story. Every tool, the stdio bridge and the skills are in docs/MCP.md.

Models

Every model runs through audio.cpp, a native runtime, so there is no Python to set up. Verified end to end and on by default:

VibeVoice 1.5B, MiDashengLM-Gen, HTDemucs 6-stem and Chatterbox Turbo have also been run end to end; the rest are marked unverified until someone reports. The 19 added in October 2026 need audio.cpp 0.9.2 or newer (soundbox model install audiocpp fetches the latest).

Any model whose licence could not be established is treated as non-commercial. This is a summary, not legal advice: read each licence before you ship what a model made.

Compared with a hosted service

Run from source

Requires Node 22.12 or later. No Python, no CUDA toolkit.

git clone https://github.com/HamSanders/soundbox
cd soundbox
npm install
npm run dev

This starts the server on port 17494 and the UI on http://localhost:5174. npm run dev:mock runs the UI with no GPU or server (see CONTRIBUTING.md).

Docker. The server and web UI run headless in one container with the audio.cpp runtime inside. CPU by default; add the CUDA override for a GPU. Both pull the prebuilt image from GHCR (:latest, :cuda, or a version such as :v0.4.0 and :v0.4.0-cuda).

docker compose up -d
docker compose -f docker-compose.yml -f docker-compose.cuda.yml up -d

To build from the checkout instead, add -f docker-compose.build.yml (and -f docker-compose.build-cuda.yml for CUDA) and --build; see docs/INSTALL.md.

There is no authentication, so keep the port on localhost or behind a reverse proxy that has some, and set SOUNDBOX_ALLOWED_HOSTS to the name you reach it by. Details are in docs/INSTALL.md.

HTTP API. Everything the app does is available on port 17494:

curl -X POST http://127.0.0.1:17494/api/generate \
  -H 'Content-Type: application/json' \
  -d '{"model":"audiocpp-stable-audio-3-sfx","prompt":"heavy wooden door slamming in a stone corridor","duration_s":3}'

Long jobs return an id with a server-sent event stream. Every route is in docs/API.md.

Docs

FAQ

Can I sell what I make? It depends on the model. Stable Audio is free under $1M revenue, ACE-Step 1.5 is MIT, and models with an unestablished licence are marked non-commercial. Each model's terms are on its card in the app.

Do I need a GPU? No, but CPU generation is slow. The hardware table above says what to expect.

Does it work on a Mac? There are installers for Apple silicon and Intel, and audio.cpp runs on Metal. None of the verified models has been confirmed on a Mac yet, so reports are welcome.

How do I remove it? Delete the app and ~/.soundbox. There is no account to close and nothing was uploaded.

Licence

Soundbox is MIT. The model weights are not: each carries its own licence, and what you can do with generated audio is governed by the licence of the model that made it. Clone only voices you have permission to use. Soundbox downloads weights at your direction and does not redistribute them. Example media under site/public/ was generated with third-party models; its provenance is in ASSETS.md.