pull/this/repo
RepoMusic and audio#sound-effects#ai-music#local-ai#mcp

Soundbox, now Mirli: free sound effects, music and songs made on your own computer

By Pull This RepoPosted 4 min read

Short answer

Soundbox, renamed Mirli in November 2026, is a free, MIT-licensed desktop app that makes sound effects, music and songs with lyrics from open models on your own computer. You type what you want to hear or paste a lyric sheet, then arrange the results on a multi-track timeline. It needs no account or credits, but it is built for NVIDIA graphics cards.

Soundbox is a free, open source studio that generates sound effects, music and full songs with lyrics on your own computer, then lets you arrange them on a timeline. Since our video it has been renamed Mirli, with a new website at mirli.app, though the GitHub repo and installer files still carry the Soundbox name. This page uses both names so you can find it either way.

What is Soundbox (Mirli)?

It is a desktop app for Windows and macOS, also runnable from source on Linux, that puts open-weight audio models behind one interface. The README describes it as "the free, open-source, local alternative to ElevenLabs and Suno". The app is MIT licensed; GitHub's API shows no licence because each model carries its own terms. The repo was created on 13 October 2026 and had 225 stars when we checked on 18 November 2026. The latest release is v0.5.0, published on 18 November.

What it makes and does, from the README:

  • Sound effects and ambience from a description, for example a window pane breaking or a thunderstorm loop
  • Instrumentals, or songs from a lyric sheet, with ACE-Step singing the words
  • Speech, voice cloning and voice conversion (only clone voices you have permission to use)
  • A multi-track timeline with effects, MIDI and mixdown, so voice lines, a music bed and effects sit in one project
  • No account, no credits, offline once the models are downloaded; everything stays under ~/.soundbox

Four models are verified end to end and on by default, and 61 more can be switched on from the Models page.

How do you install Soundbox?

There is no single install command. Download the installer for your system from the download page or the latest release: Soundbox_<version>_x64-setup.exe for Windows 10 and 11, or a .dmg for Apple silicon or Intel Macs. The installer carries its own runtime, so there is no Python or CUDA toolkit to set up.

The builds are not signed yet, so the first launch asks you to confirm. On Windows, choose More info, then Run anyway. On a Mac, open System Settings, Privacy & Security and choose Open Anyway. If macOS calls the app damaged, the project's docs give this fix:

terminal
xattr -dr com.apple.quarantine /Applications/Soundbox.app

Then open SFX, pick a model and generate. The first time you pick a model, it shows the licence and downloads the weights. On Linux or a server, the README's route is to run from source with Node 22.12 or later:

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

The web UI has no login, so keep its port on your own machine.

What hardware does it need?

Your machineWhat to expect (README)
NVIDIA graphics cardCUDA; a few GB of video memory covers the verified models, MiniMax Music 3 wants about 10 GB
AMD or Intel graphicsruns on the CPU unless you set SOUNDBOX_AUDIOCPP_BACKEND=vulkan, which is not yet confirmed
Apple silicon MacMetal; installers exist, but no verified model is confirmed on a Mac yet
No dedicated GPUworks, slowly: seconds on a GPU become minutes on the CPU

What prompts work best?

The project's rule is to describe the material, the action and the space; mood words do little.

Instead ofWrite
glass breaklarge window pane breaking, cracking then collapsing inward with long shards falling and shattering on a wooden floor
laserplasma bolt firing, thick electric discharge with a downward pitch sweep and a crackling tail
retro music1980s synthwave arcade title theme, wide analog saw lead, gated reverb drum machine, pulsing sidechained bass, 110 BPM

A few more habits from its docs: write a sentence rather than a keyword list, keep one sound per prompt and combine them on the timeline, and for songs keep lyrics short and repeated. Mark sections with [verse] and [chorus]. Lock the seed while you tweak wording, since the same seed, prompt and model give the same audio.

The song in our video came from this lyric sheet on ACE-Step 1.5 Turbo, with the style prompt "triumphant power pop victory anthem":

text
[chorus]
We took the long way home
We took the crown
Victory lap, victory lap
We are not coming down

The samples site has around 600 more prompts with the sound each one made.

What is the catch?

  • It wants an NVIDIA card. Without one it still works, but a clip can take minutes.
  • Unsigned installers, so Windows and macOS warn on first launch.
  • Mac support is unconfirmed for the verified models.
  • Commercial use depends on the model, not the app. Read each model's licence card.
  • It is young and changing: a first release in October 2026 and a rename in November.

Who is Soundbox for?

Game developers, video editors and podcasters who need lots of sound effects and beds without a subscription, people who want to try open music models without wiring up their own workflows, and developers who want an agent to handle a sound pass through MCP. If you want polished songs in a browser with no hardware at all, a hosted service is simpler; our video compared it with Suno Pro at $10 a month on Suno's US pricing page.

How does it compare with other free audio tools?

ToolWhat it is
Soundbox (Mirli)one desktop app with effects, music, songs, speech and a timeline
ACE-Step 1.5the open song model Soundbox uses for vocals, on its own (MIT)
YuEopen model that writes full songs with vocals from lyrics (Apache-2.0)
AudioCraftMeta's library behind MusicGen and AudioGen (MIT code)

What did we run for the video?

We did not install Soundbox. The video used the authors' 30-second tour video, README screenshots, and the project's own published samples of a breaking window and the "Victory lap" chorus, played as they are. For more free music tools, see LYNX SHARE, a beat maker that opens FL Studio projects, and Honeydew, a browser remake of FL Studio Mobile.

Pull it. Stash it.

Questions people ask

Is Soundbox the same as Mirli?

Yes. The README says Mirli was formerly Soundbox. The GitHub repository, the soundbox command, the MCP tool names and the installer file names keep the old name for now, and the website moved to mirli.app.

Can I use sounds made with Soundbox commercially?

The app is MIT, but each model has its own licence, shown before you download it. Per the project's notes, ACE-Step 1.5 Turbo is MIT, the Stable Audio 3.0 models are free under $1M revenue under the Stability AI Community licence, and models with an unclear licence are treated as non-commercial. Check each licence before you ship; this is not legal advice.

Does Soundbox work without a graphics card?

Yes, but slowly. The README says a clip that takes seconds on a GPU takes minutes on the CPU. NVIDIA cards run on CUDA; AMD and Intel cards fall back to the CPU unless you switch on the Vulkan backend, which is not yet confirmed on the verified models.

Does Soundbox work on a Mac?

There are installers for Apple silicon and Intel Macs, and the audio engine runs on Metal, but the README says none of the verified models has been confirmed on a Mac yet.

Can Claude Code or Cursor use Soundbox?

Yes. It runs a local HTTP API on port 17494 and an MCP server, which the README now lists with 24 tools, so an agent can generate sounds and place them for a game or video.

Sources

  1. 1Soundbox (Mirli) repository and README
  2. 2Mirli website and download page
  3. 3Mirli sound effect samples
  4. 4Soundbox latest release (v0.5.0, 2026-11-18)
  5. 5Soundbox desktop install notes (docs/DESKTOP.md)
  6. 6Soundbox prompt guidance (docs/GENERATING.md)
  7. 7GitHub API: HamSanders/soundbox (stars, language, created date, checked 2026-11-18)
  8. 8Suno pricing in US dollars (read 2026-11-15)
  9. 9ACE-Step 1.5
  10. 10YuE
  11. 11AudioCraft

Get tomorrow's repo in your feed

One free, open source repo a day as a short video, shown working, with the catch named. Pull it. Stash it.