Do I need an NVIDIA card?

No. The CPU backend is always included and is the reference the CUDA path is tested against. An NVIDIA GPU adds the CUDA path.

Is 128 layers a marketing number?

It is 128 logical layer slots in eight groups, and all of them can be loaded. It is not a promise of 128 simultaneous voices — real polyphony depends on the patch, the backend and your hardware.

Does it run on macOS?

No. Windows and the Debian family including Ubuntu.

What does AI Arrange actually do?

A language model plans an arrangement; deterministic code builds it; every step is undoable. It runs locally via Ollama or through an OpenAI-compatible provider you configure. It is a preview.

Can I use it as a plugin in another DAW?

Plugin export exists in the codebase but is not part of the Founder Beta offer. It is not something to buy on today.

What happens to my €49 at 1.0?

The Founder terms are being finalised in legal review and are shown to you before any payment. Nothing is charged when you apply.

VibeField Mosayac inside VibeField Studio

Build the sound. Then build the track around it.

128 layers, spectral mixing, CPU included and NVIDIA CUDA optional — with arranger, automation and export in the same flow.

Apply for Founder Beta Watch the 60-second demo

8 groups of 16 layers — the structure Mosayac actually uses.

One layer, one group, the whole patch

The same patch at three depths. Nothing here is a rendered impression of a synth — it is the instrument doing what the layer architecture is for.

Specifications and limits

Verified

8 groups of 16 layer slots, stably addressed.
Per layer: gain, pan, transpose, fine tune, ADSR, HP/LP, four-band EQ and up to eight unison voices.
Batch edit, distribute, deterministic randomise, copy/paste and group mute, solo and transpose.
The same engine runs on CPU and on CUDA; the backend is a choice, not a separate edition.
Spectral processing places layers in the mix instead of stacking them.

Limits

128 counts logical layer slots, not 128 guaranteed simultaneous voices.
Real performance depends on the patch, the polyphony, the backend and your hardware.
CUDA needs a supported NVIDIA GPU and a matching driver. This is not a Vulkan, Metal or AMD claim.
CPU and CUDA are numerically very close but not bit-identical.
We publish no speed-up percentages until reproducible benchmarks exist for named hardware.

128 parts, one instrument

Mosayac organises up to 128 layers in eight musically usable groups. Each layer can take its own character, space and spectral position; the groups turn the mass back into something you can play.

Specifications and limits

Verified

8 groups of 16 layer slots, stably addressed.
Per layer: gain, pan, transpose, fine tune, ADSR, HP/LP, four-band EQ and up to eight unison voices.
Batch edit, distribute, deterministic randomise, copy/paste and group mute, solo and transpose.
Wavetable engine with five built-ins, mip levels, morph and frame scan, and imported multi-frame WAVs.
Additive engine with 1–64 partials, tilt, odd/even balance, inharmonicity and phase spread.
Deterministic noise with a colour filter, plus two-operator phase FM with feedback per layer.
Modulation matrix with LFOs and random walks, performance macros, per-layer shaping EQ and a shared reverb send driven by the Space macro.

Limits

128 counts logical layer slots, not 128 guaranteed simultaneous voices.
Real performance depends on the patch, the polyphony, the backend and your hardware.
Wavetable import is WAV-based in the current workflow.
There is no visual wavetable drawing editor.
Per-voice modulation sources currently use the most recent active voice as a proxy.
This is not MPE and not an unlimited modulator count.

Explore Mosayac →

Room in the spectrum

Spectral mixing gives layers their own place instead of stacking them until the mix turns to mud. What you hear is the same patch before and after — no loudness trick.

Specifications and limits

Verified

The same engine runs on CPU and on CUDA; the backend is a choice, not a separate edition.
Spectral processing places layers in the mix instead of stacking them.

Limits

CUDA needs a supported NVIDIA GPU and a matching driver. This is not a Vulkan, Metal or AMD claim.
CPU and CUDA are numerically very close but not bit-identical.
We publish no speed-up percentages until reproducible benchmarks exist for named hardware.

CPU is included. CUDA is added.

There are no CPU and GPU editions. The CUDA build contains the CPU backend as well; an NVIDIA card is only required for the CUDA path itself.

Specifications and limits

Limited

The same engine runs on CPU and on CUDA; the backend is a choice, not a separate edition.
Spectral processing places layers in the mix instead of stacking them.
Windows bundles and .deb packages for the Debian family, with a four-language interface.

Limits

CUDA needs a supported NVIDIA GPU and a matching driver. This is not a Vulkan, Metal or AMD claim.
CPU and CUDA are numerically very close but not bit-identical.
We publish no speed-up percentages until reproducible benchmarks exist for named hardware.
Windows builds are not code-signed yet, so SmartScreen may warn.
No macOS package, and no AU or AAX support.
Support is stated per distribution and version from the test matrix, not as "all Debian-based".

From patch to track

Mosayac, clip, arranger, automation and mixer, then freeze, stems and WAV — one project, no export dance between tools.

Specifications and limits

Verified

Sound design and arrangement live in one project and one file format.
Tracks, clips, scenes and a timeline with tempo and meter changes.
Automation lanes for track volume and pan, device parameters and synth parameters, applied identically live and on bounce.
Freeze tracks, bounce the mix or per-track stems, with a bit-identical CPU render.

Limits

This is a focused production arc, not a replacement for a general-purpose DAW.
There is no macOS package.
No audio comping and no take lanes.
No claim of feature parity with long-established full DAWs.
No live automation recording in touch, latch or write modes.
Automation resolution differs by target; it is not uniformly sample-accurate.
Export is WAV. No MP3, no dithering, no loudness normalisation and no mastering chain.

See the Studio workflow →

AI Arrange, with limits

The model plans, deterministic code builds, and undo keeps you in control. It runs locally through Ollama, or through an OpenAI-compatible provider if you choose one.

Specifications and limits

Preview

A language model produces an arrangement plan; deterministic code renders it into scenes, clips and notes as one undoable step.
Runs locally through Ollama, or through an OpenAI-compatible provider you configure.

Limits

A preview, not an autonomous producer and not a quality guarantee.
Privacy depends on the provider: local stays local, a cloud provider receives your data.
The current prompt is focused on psytrance and electronic arrangements.

Built to be checked

CPU and CUDA are tested against each other, projects are written atomically, and the test snapshot is dated rather than implied.

Specifications and limits

Verified

An automated test suite covers the audio, synthesis, project and DAW cores.
Atomic saves, autosave with recovery and a versioned project schema with migrations.

Limits

The test count is a dated snapshot of one build, not a permanent property.
No claim of being crash-free, latency-free or guaranteed real-time on any hardware.
No cloud backup and no collaborative multi-user editing.
Recovery is not a substitute for your own backups.

Platforms and honest limits

Windows and the Debian family including Ubuntu. There is no macOS build, no finished live launcher, no audio input recording and no MPE — and we would rather say so here than let you find out after paying.

Specifications and limits

Limited

Windows bundles and .deb packages for the Debian family, with a four-language interface.

Limits

Windows builds are not code-signed yet, so SmartScreen may warn.
No macOS package, and no AU or AAX support.
Support is stated per distribution and version from the test matrix, not as "all Debian-based".

Check system requirements →

Questions

Do I need an NVIDIA card?: No. The CPU backend is always included and is the reference the CUDA path is tested against. An NVIDIA GPU adds the CUDA path.
Is 128 layers a marketing number?: It is 128 logical layer slots in eight groups, and all of them can be loaded. It is not a promise of 128 simultaneous voices — real polyphony depends on the patch, the backend and your hardware.
Does it run on macOS?: No. Windows and the Debian family including Ubuntu.
What does AI Arrange actually do?: A language model plans an arrangement; deterministic code builds it; every step is undoable. It runs locally via Ollama or through an OpenAI-compatible provider you configure. It is a preview.
Can I use it as a plugin in another DAW?: Plugin export exists in the codebase but is not part of the Founder Beta offer. It is not something to buy on today.
What happens to my €49 at 1.0?: The Founder terms are being finalised in legal review and are shown to you before any payment. Nothing is charged when you apply.

Founder Beta

A small, qualified group that gets in early, shapes priorities and accepts that a beta is a beta.

€49 after acceptanceNothing is charged when you apply. If your application is accepted you receive the terms first, and only then a payment link.

Apply for Founder Beta Check system requirements