The Ableton AI assistant
that measures while you play

What does an Ableton AI assistant actually observe?

The phrase covers products that share a chat box and almost nothing else. The thing that separates them is not the model behind the box. It is the input.

An assistant can observe a manual, a project file, one rendered bounce you upload, the running session, the screen, or the audio actually coming out of your machine. Those are six different inputs and they answer six different sets of questions. A tool that reads a manual cannot tell you your bass is wide. A tool that reads a project file cannot hear that the master is clipping. A tool that scores an upload heard one bounce, once, and not the loop you are on.

So the useful question is not which one is smartest. It is: what is it looking at, and what does it change. There is a longer answer to that in what this category actually is.

What does Miles measure while your set is playing?

Miles taps the audio leaving your machine and measures it twice a second, while it plays. Nothing is bounced, uploaded or queued.

• Loudness, in LUFS, integrated and short term.
• True peak.
• Key and chord, but only while what is playing is a loop. In silence no number is written at all.

The measurement runs locally. The audio itself does not leave your machine. What travels with your question is a numeric summary of it, and what that summary is worth is a thing we publish rather than assert: here is one of those measurements with its limits attached.

A missing field means we do not know. It never means a default. That rule is in the contract the app and the backend share, not in a style guide, because a made up default is how a measurement tool starts lying quietly.

What does Miles read from your set, and what does it never read?

Miles connects to the open set through a read only bridge. The scope is listed here in full, because a scope that is only described is a scope that grows.

It reads: track names, the devices on each track, which track and which device you have selected, the selected device parameters, the tempo, and whether the set is playing.

It does not read: faders, mutes, solos, sends, automation, meters or clip fades.

It writes: nothing. There is no path from Miles into your set. Not a clip, not a device, not a parameter.

What does it see on your screen?

One still of your DAW window, taken at the moment you ask a question that needs eyes. There is no video and no continuous capture.

We measured this rather than described it. With the eye on, one question produced a request body of 327,877 bytes, and 218,911 of those bytes were a single 1200 by 792 PNG. With the eye off, the same question in the same session produced 31,763 bytes and no image field at all. Not an empty field: no field.

Sitting idle for 120 seconds with the app and the set open produced zero open sockets across twelve samples, and zero requests.

The full byte level run is published, including the sentence in our own privacy copy that the measurement proved wrong. The complete scope of what travels is on the privacy page.

Does it change anything in your project?

No, and this is the line the category splits on.

Miles does not write to your set, does not insert devices, does not move parameters, does not render audio and does not touch your files. It measures what is already happening and tells you what it reads.

That is a smaller promise than fixing your mix, and it is the reason the answers can be checked. A number you can see in your own meters is a number you can argue with.

Why not just ask a general purpose chat assistant?

Because a general purpose assistant is answering from a recipe, not from your session. It cannot hear the loop that is playing, it does not know the tempo you are on, and it does not know which of the devices on your screen you have selected. Ask it why your low end is not translating and it will describe low end in general, correctly, and about a track it has never heard.

Ask Miles the same thing while the loop plays and the answer starts from numbers taken off that loop, half a second ago. Where the other tools in this category sit, named and with their own words quoted, is a separate page.

What Miles cannot do

• It cannot change your mix. It has no write path into your set and no audio processing of its own.
• It cannot generate MIDI, load an instrument or build a drum pattern.
• It cannot separate stems.
• It cannot tell you the key of something that is not playing.
• It cannot see anything outside the DAW window it captures, and it does not read your project file.
• It does not read your faders, sends or automation, so it cannot tell you what you did, only what came out.
• It cannot recommend a plugin you do not own, because the only inventory it has is the one you scanned yourself.

Which Ableton version has this actually been run in?

Ableton Live Live 12, on a real machine, in a real session, with the window open and the track names read back one by one. That run is the only reason Ableton Live is named on this site at all.

Live 11 is untested here. Live Lite and Intro are untested here. We are not going to guess on your behalf: a version that has not been run is a version we have nothing to say about.

Miles connects to the session on its own. There is nothing to install inside Ableton Live, no Max device and no Remote Script for you to place by hand.

Questions