Privacy claims need a data-flow diagram behind them
“Private” can describe several different architectures. Audio may remain entirely on the device, travel to a cloud model, be stored as a recording, or exist only long enough to stream. Those choices are not interchangeable, so a useful privacy page should say exactly which one applies.
Quill uses cloud transcription but does not create a local recording archive. Microphone samples are buffered in memory, streamed from the Mac app to Gemini, and released when the live session ends.
Memory-only audio limits what the app can retain
Volatile memory is used for the small buffers needed to convert and send live audio reliably. Quill does not write those buffers to an audio file. When dictation completes, is cancelled, or fails, the session disconnects and the pending buffers are cleared.
That means Quill cannot offer playback of old dictations—and cannot expose a library of old voice recordings—because that library does not exist.
Bring your own key removes an extra account layer
A bring-your-own-key app asks you to authenticate directly with the model provider. Quill does not operate an account system or a proxy transcription server. Your Gemini API key is stored locally with owner-only file permissions and used by the Mac app to open the live session.
BYOK does not make cloud processing local. Audio still reaches Google for transcription, and Google’s API terms and data policies remain relevant. What it removes is an additional Quill-controlled server and credential database.
Transcript history and usage counts are different data
A transcript history preserves the content of what you said. A usage counter preserves a number. Quill stores daily and lifetime dictation and word counts locally, but not the text behind those counts. You can reset the numbers in Privacy & Access.
The latest transcript stays in memory while Quill runs so Paste Last Transcript can work. It disappears when the app quits and is not added to a searchable archive.
Questions to ask any voice typing app
The best privacy comparison starts with concrete questions rather than a badge. Ask where audio is processed, whether recordings are written to disk, whether transcript history is enabled, what account is required, what permissions are used, and whether the implementation can be inspected.
- Is microphone audio stored locally or remotely?
- Does the app retain transcript content or only aggregate counts?
- Does audio pass through the developer’s own server?
- Can the API credential and local data be removed or reset?
- Is the client source available for inspection?