Skip to content

Questions

The questions a careful engineer asks before letting a new tool near a desk, with short answers and a link to the page that says more.

Here is exactly where AI is used, and what holds it in check:

  • A model drafts the plan from your input list and conventions, on your laptop, before the show.
  • Code with no model in it checks every value in the draft against the desk’s real limits: sockets this model has, stereo pairs it can form, values in range. A setting the desk cannot hold is an error, and nothing is written.
  • Every write is read back from the desk, and anything that did not land is named.
  • Every choice the tool made that your files did not state is announced as an INFO finding, with what it was decided from (Reading findings).

Most of the code was written with AI coding agents, to the design and under the review of a working live-sound engineer, and tested against a real WING Compact. There are about 4,500 automated tests. Read the plan it produces, on the Demo show, and judge that.

No. The model is never connected to the desk: it writes a file, before the show, and ordinary code applies it after you have read it. Watch, the part meant for a show, only reads, and uses no model.

A setting the desk cannot physically have is refused before anything is written. A choice that is possible but that you did not state, such as a bus being mono or stereo, is made and printed as a note, so you see it in the plan rather than at soundcheck. The model is told to set an EQ band, a gain or a trim only where your conventions state one; your sound otherwise comes from your own channel presets. The optional critic pass flags a value that appears nowhere in your files.

Because your rules stay in your own English, not in a format you have to learn. Reading “drums go to DCA 1 unless it is a festival changeover” is what the model is for. Verify, apply, undo, audit, watch and the Demo show run with no model.

Your input list and conventions go to the model you set up, with your own key: Claude, which Atoo is built and tested on. A local model through Ollama keeps them on your own machine, but local models have not worked well so far (Setting up an AI provider). There is no Atoo account, and nothing is sent anywhere else.

Your AI provider bills each build to your key. Each call to the model shows what it cost on Build step 04 while the build runs, and the build record in the show’s builds/ folder keeps it. Two recorded builds of a 38-channel festival show cost about $0.35 and $0.60 on Claude, with the critic pass on, using an earlier default model and effort. A smaller show costs less. Verify, apply, undo, audit, watch and the Demo show cost nothing: they use no model.

A full apply takes the whole desk: it first puts the desk back to its initialized state, leaving the network, clock, cards and monitor listening alone, then writes the plan and reads it back. Save your show first (Apply a plan). A patch adds or changes only what you typed and proves nothing else moved (Patch a show). Nothing is written until you tick Arm on the Apply screen and press its write button.

Every write is recorded with the values it replaced. Undo puts back the newest one, address by address, and leaves alone anything changed since (Undoing a write). A saved show on the desk is always a reload away.

Nothing to engineers. It is free and open source, to use on any show, paid gigs included, with no account to open with us (a build uses your own AI provider’s key). A company that wants to build it into a closed or commercial product can get a commercial license by agreement. It was written by a working live-sound engineer and consultant for real shows.

Its author uses it on real shows, which is why it exists. It is open source: the code, its tests and these pages are in its repository, so someone else can pick it up.

Behringer WING only: the Compact, the full-size WING and the Rack. No other console, the X32 and M32 included.

No. Nothing moves a fader during a show. It sets the desk’s starting state and gets out of the way.

Why Atoo? compares it with building by hand or in WING Edit, copying last month’s show, the desk’s presets, and a spreadsheet input list.

Open an issue on the repository with what you ran, what you expected, and what you saw. Reporting a problem says what to send, and which files help.

Atoo 0.2.6