Learn
Offline gets used for two different claims. Here is the line between them, step by step.
Offline means the audio to text step happens on the device doing the transcribing, with no server in that loop. It does not mean the app never touches a network: a model download, a license check, or an update can still need one, just not on every recording.
Offline gets applied to two different claims. One is that the app was set up without a network at all. The other, far more common, is that the app runs the transcription step itself without a network, once it has what it needs on the device. A model download during setup, license activation and its periodic refresh, or an update check are all still network events. None of them are the same as the recording itself touching a server.
Four steps, checked against what actually needs a connection and how often. The question that matters is whether the transcription step itself depends on a network, and whether the app is clear about when the other steps go online.
On a local architecture, the audio file stays on the device it was recorded or imported on for the whole transcription step. It is not copied to a server to be processed and copied back. What happens to the file afterward, kept, exported, or deleted, is set by the app's own retention choice, not by the network.
Where a transcription product ships a local API alongside its desktop app, a script or an agent can call the same on-device model and get the same transcript the desktop app would produce, with the audio staying on the machine either way. EdgeSpeak's local API works this way: the desktop app and a script or an agent calling it directly share the same local transcription engine.
EdgeSpeak transcribes with its local transcription engine on your own computer. The model downloads once; after that, recordings and transcripts stay on the device. The license is activated online and refreshed online from time to time, and none of that traffic carries audio or text. With a purchased license you can keep transcribing offline during the license grace period; once it runs out, go online once to refresh it. The free trial is different: it checks in online each time the app starts. The desktop app supports Apple Silicon Macs and Windows 10/11 x64 (the Windows installer is not yet code-signed); on Linux x86_64, the edgespeak-cli command line (CPU or CUDA) runs the same local engine. Every device can start with a 7-day free trial.
Most do, for specific tasks, but not for every recording. A one-time model download, an occasional license check, and an update check are the typical uses of the network. The transcription step itself does not need one while the model is on the device and the license is valid.
It depends on the model, not on whether the model runs locally. Accuracy varies with audio quality, accent, and language wherever the model runs, so test any app on one of your own recordings rather than relying on a marketing number.
Where the product ships a local API alongside the desktop app, yes. The same on-device model answers both the desktop app and a script or an agent calling it directly.
On a local architecture, the audio stays on the device it was recorded or imported on. Whether the file is kept, exported, or deleted afterward is the app's own retention setting, not something the network decides.
