IcedAudio
The Iced Audio promise shield: a Viking round shield with the Iced Audio wave painted across it Our policy

Our promise: AudioCortex is not an AI music generator.

Many AI music generators are trained on other artists’ songs without permission and make variations of their music. AudioCortex doesn’t do that, and never will. No AI makes the audio: every sound comes from your recordings or from the synthesizers and effects you control.

— Iced Audio

We’re artists too

Our own music is probably in the training data of AI music generators. We understand how complicated it is to train models on people’s music, and that’s why we don’t do it, and why we won’t build any generative music service or API into AudioCortex.

Our goal is to use AI for one thing: taking over the repetitive tasks, so you’re free to be creative.

AI has been in AudioCortex since July 23, 2018

We’re not new to this. AudioCortex began on July 23, 2018, as a sound browser that used machine learning to classify your sounds, so you could find them by what they sound like. From day one, AI has been a tool for organizing and controlling your library, never for making the music.

Designed for artists’ rights

Over the years, our customers have been the people who make the sound in the music industry: film and TV soundtracks, sound design, game audio, and artists. We support artists’ rights, and that is the design philosophy of AudioCortex.

It’s why AudioCortex has its own local model, so you can automate the basics entirely on your own computer, or plug in your favorite frontier model if you prefer.

What the AI does: control and automation

Oscilla, the assistant in AudioCortex, operates the app for you. Ask it to find a lush guitar, stretch it into a cathedral, lay it on a track and bounce it, and it runs the same tools you would, step by step, in front of you. It wires patches in Nodes, sets up the drum machine, builds an arrangement on the built-in instruments, and mixes. You can watch every step, change it, or do it yourself.

Bring any model

You choose what drives Oscilla:

  • Oscilla Local: Iced Audio’s own model, a free 2.5 GB download that runs entirely on your Mac, with no account and no key.
  • Claude: your own Anthropic key, or sign in with your Claude account.
  • Gemini: your own Gemini key.
  • OpenAI: GPT with your own OpenAI key.
  • Your own model server: LM Studio, Ollama, llama.cpp, or anything that speaks the OpenAI format, on this Mac or another machine on your network.
  • Apple’s on-device model: on macOS 26 with Apple Intelligence, nothing to download.

The library, search, tagging, texture engines, Nodes and instruments all work without an assistant too.

Where every sound comes from

  • Your recordings: the sounds in your library.
  • The built-in instruments: wavetable, analog and FM synths, a drum machine and samplers.
  • The texture engines and effects: XXL Stretch, Mosaic, convolution, the spectral suite and more.

All of it is signal processing that you control. No diffusion, no text-to-audio, no AI-made samples.

Your sounds stay yours

Listening, tagging and search happen on your Mac, and your audio never leaves it. The only thing that ever goes out is Oscilla’s text conversation, and only to the provider you chose. With Oscilla Local, nothing goes out at all.