fAIble Labs
How it works

The machine runs the exhaustive part. The engineer makes the call.

Taking an unknown recording to a characterization is senior work, and most of it is working through possibilities and dead ends. That is the part the software does. What it hands back is a short, honest list.

One recording, five steps

IntakeA raw recording is placed on the box and verified. It never leaves the hardware it was put on.
TriageThe recording is sorted by the kinds of signal it holds, so the right instruments run next.
CharacterizeStructure, timing, framing and modulation family are measured, each with its own controls.
AttributePulses, hops and carriers are linked to the radio that sent them, with the evidence stated.
AdjudicateThe engineer gets a short list to accept, correct or mark unsure. The tool never decides for them.

Honest by construction

Every check has three answers, never two.

Yes

The check looked, and the instrument showed it could see.

No

The check looked and the answer was no. A recorded no is a result.

Could not look

The check could not run on this recording. It says so, and that holds the result. It is never reported as a pass.

The same rule covers attribution. Two radios sharing a channel are never called one radio because they look alike; “indistinguishable” is a missing measurement, not a verdict. When the software says one radio, it says which test showed it and how far it could go.

Sovereign by design

  • On the box. Recordings, analysis and results live on the hardware they were put on. The data is yours.
  • No telemetry. Nothing is reported to us. Licensing is a signed key read locally.
  • The model, stated. The assistant can run hosted, on a model you already operate, or entirely on the box; the software is designed for all three and we will tell you which your installation uses.
  • Numbers from code. Every measurement comes from signal-processing code with its own controls. The model proposes and explains; it never produces a number.

What we never do

  • Decode or read the content of any transmission. The product works from the physical footprint only.
  • Decide what a signal is for classification or policy purposes. AI flags; people decide.
  • Move your data. Not to us, not to a cloud, not to a model outside your boundary without your installation being set up that way on purpose.
  • Report a result the instrument did not earn. An empty result has to prove the instrument could see something known to be there; otherwise it reads “could not look”.

Start with one recording

Unclassified, one you own or are authorized to hold. Tell us what you need to know about it.

Contact fAIble Labs