Why “Never Records” is an invariant, not a slogan.

Most AI note tools market deletion. Fazit was built so that deletion is unnecessary.

The problem with “we delete it later”

Every cloud-based meeting assistant necessarily writes audio (or a processed representation) to disk or object storage. This is required for transcription, diarization, model improvement, and compliance logging.

Once data is written, it can be subpoenaed, breached, or used for training. Even if the vendor promises deletion, the file existed.

How Fazit is different by construction

• Audio is captured through a Core Audio process tap into a fixed-size ring buffer in RAM only.
• Transcription and note generation consume a snapshot.
• AudioRingBuffer.reset() is called unconditionally on every exit path.
• The type has no serialization method. It is impossible for the current implementation to persist audio without a deliberate, reviewable change.

This is not a policy. It is an invariant enforced in code — and source access for independent security review is available on request.

Legal significance (Germany §201 StGB)

German counsel has noted that a pipeline that transcribes in volatile memory and never writes an audio file to a sound carrier generally does not meet the elements of the statute. The distinction between “never created” and “created then deleted” matters in law.

Why competitors struggle to copy this

Cloud vendors built their economics on storing data and training models. Re-architecting to RAM-only destroys their current value proposition. Local competitors can add it, but most still buffer to disk for reliability and features.

FAQ

What does never records actually mean?

That no audio file is created at any point. Audio is captured into a fixed-size ring buffer in RAM, transcription consumes a snapshot, and the buffer is reset unconditionally on every exit path including errors. There is no code path that writes audio to disk, which is why every note can carry audio_retained: false.

Is never recorded the same as deleted after transcription?

No, and the difference is the whole argument. Deletion is a promise about future behaviour, enforced by process, auditable only through a report you have to request. Never created is a property of the execution path. A subpoena, a breach or a change of terms only has to find the window where the file existed.

Can I verify the audio is never written to disk?

Source access for independent review is available on request for security teams. Short of that, the observable signals are that no audio file appears anywhere on the machine and that the app requests the audio-capture permission rather than screen recording.

What is stored, if not the audio?

One Markdown file per call, written into the folder you choose in your Obsidian vault, plus small application settings. There is no app database and no server-side copy of the note.