Two things from OVNI Audio today.

EYEPIECE 0.1.0 is new, and it's not a plug-in: a free Mac app that reads your project file and writes a snapshot of it in text, for whatever AI you already use. The AI gives the advice; EYEPIECE only reads.

What it reads, honestly:
- Ableton Live in depth — tracks, clips, notes, devices, third-party plugins and the mix (verified up to Live 12.3.7).
- Logic, GarageBand and FL Studio at their first level only — tempo, the outline of the project, the plugins by name in FL. Not the notes or the mix.
- Not the audio. Every number comes from the project file.

Each finding comes with its number and the id of the rule that produced it, and the snapshot also lists what it did not read, each with its reason. No internet, no telemetry, no AI inside. Free and closed-source, macOS 11+.
https://ovniaudio.com/eyepiece

TELESCOPE 0.2.0 — the free, open-source analyser (AGPLv3, VST3/AU on macOS, VST3 on Windows). The main fix: an open TELESCOPE window no longer stalls the DAW. Also new: a light theme, readable axis labels, VERDICT at 125 % and 150 %, and a section of the reference for TONAL BALANCE. What a file measures didn't change.
https://ovniaudio.com/telescope

Happy to answer questions here.

 

Oh great another way for AI to steal artists' work.

 

what a bad idea

 

How so? Its not a generative AI.

Seems pretty useful to me. I often work in sessions well over 100 tracks, having an app like this that reports a detailed overview of your session is a welcome addition.

 

It says it doesn't require internet. It's says from what I can glean from the gobble-d-gook lingo on the web page that "AI" is involved. Then up top here it says no "AI inside".... Sounds like snake-oil to me... Seems a bit redundant IMO since most of the tools/metering/processors/monitoring we all use tell us what we need to know and doesn't involve "AI" (or shouldn't at least).

 

If I had too much time I would have a look into the network traffic that thing generates.
Wouldn't it be a great feature if it trained it's daddies servers at home?

 

From what I can see, this gives you a much more micro level report than the audio analysis tools we are used to. You use your own AI. Both things are true. There is no "AI inside", but it requires an external AI to operate. Meaning no requests run through the developer, only through whichever model you use. I haven't checked, but I would imagine you can use this with even 100% local models.

 

Claude Code reads and writes Ableton Live files anyway (just not the blobs).

 

It looks a lot less useful than controversial. How we can go from it telling you some samples are missing from a project, to a full blown AI conspiracy theory is pretty amusing.

 

Thanks for the replies. A few facts, so nobody has to guess:
- EYEPIECE makes no network requests at all: no telemetry, no server, no account. It works the same with no internet connection.
- There's no AI inside, and nothing is generated or trained. It reads your own project file, without modifying it, and writes a plain-text list of what's in it. You see all of it before you copy it. If you paste it into an AI, it goes to that AI under its rules, the same as if you had typed it yourself, and that includes local models.
- It doesn't listen to the audio, so it doesn't replace your meters. It reports the project: tracks, clips, notes, devices, third-party plugins, the mixer, and a list of what it couldn't read, each with its reason.
- @Foobar yes, an .als is gzipped XML and a coding agent can open it. EYEPIECE does the reading without an agent: read-only, with each native parameter marked verified, inferred or raw, and it also reads Logic, GarageBand and FL Studio at their first level.
If it's not for you, fair enough.

 

Can you provide some real world of examples of what AI could do with the EYEPIECE text? Or what/why would I ask AI with EYEPIECE as my context?

 

Do you plan on adding Cubase support?

 

Where do these "models" come from? Has to be either internet or "subspace relay" maybe? pffftWith regard to a couple of other apps tested lately where "AI" was said to be involved, also "no internet" was floated as well, the installer did nothing but attempt to DL a literal ton of shit from some dev server. If this is anything like those other apps then it would stand to reason I would impart the same fate to this one as I did to the others - e.g. "the bit-bucket"...

 

@mopoge a few real ones. You paste the snapshot and ask what you'd ask someone looking over your shoulder: "the chorus doesn't lift, what would you check first?", "give me an overview of this 100-track session", "which third-party plugins does this project need before I move it to another machine?". In a Live set the AI sees the tracks, devices, mixer and the shape of the arrangement instead of guessing, and if it needs more (an EQ, the notes of a clip) it asks for a code and EYEPIECE copies just that block. The advice is the AI's, and it can be wrong; what EYEPIECE adds is that it works from what's actually in the file, with a list of what wasn't read.
@clone not in 0.1: it reads Ableton Live in depth, and Logic, GarageBand and FL Studio at their first level. I can't promise other DAWs, but Cubase is noted.
@saccamano the model is whatever AI you already use, in a browser or running locally. You copy the text and paste it there yourself; EYEPIECE never talks to it. There's no installer either: it's a .dmg, you drag the app to Applications, and it downloads nothing afterwards. If you'd rather check, run it with Wi-Fi off or behind Little Snitch: it works the same.

 

@Ovni audio : Thanks! Very interesting. I suppose one could have AI "listen" to your track as well, though if I were to do that I'd want an AI that is totally offline.

How have your results been with your own AI feedback?

 

@mopoge honestly, I don't have numbers I'd stand behind: it depends on the model and on the question. What changes is what the AI has in front of it: your actual tracks, devices and sends (in a Live set) instead of a guess, plus a list of what the snapshot didn't read, so it has something to point at when it can't tell. Whether the advice is any good is still yours to judge.
On listening: EYEPIECE doesn't, and in 0.1 it doesn't read what TELESCOPE measures either. For a fully offline setup, use a local model: EYEPIECE sends nothing, so with a local model nothing leaves your machine.