Why Lven Instant uses on-device recognition
The privacy and reliability trade-off, without turning architecture into a universal benchmark claim.
On-device speech recognition changes the data path: microphone audio can be processed by the client rather than uploaded for remote inference. That reduces exposure of dictated audio to ClassEve infrastructure and removes the network round trip from the recognition stage.
It does not make every device equally fast. Completion time varies with the CPU, model state, recording length, thermal conditions, and other workload. It also does not make the whole product permanently offline: initial model setup or repair and account or entitlement validation can require a connection.
That is why ClassEve keeps Instant and Cloud as separate products. Instant prioritizes an on-device transcription path on supported hardware. Cloud sends audio to a server by design and can serve devices or use cases that prefer that trade-off. Neither path is described as universally faster or more accurate for every user.