Skip to content

Performance and memory

Model preparation, transcription, local summaries, chat, or indexing takes longer than expected or makes the device less responsive.

Source duration, model size, available memory, active applications, local runtime startup, and hardware characteristics affect local work. The catalog’s size estimates are not validated performance requirements.

Identify the current stage and model, allow one active operation to finish or fail, and compare a short safe recording with the same settings.

Choose a smaller installed local model only after testing whether its output is acceptable for the task. Avoid concurrent model downloads and other resource-heavy applications when you need a stable local run. Report reproducible behavior instead of assuming a device is supported or unsupported.

Do not force a reset during active recording, conversion, download, or generation. Preserve source recordings and exports before making destructive changes.

Collect model identifier, task type, media duration, available-storage observation, device and macOS details, elapsed time, and whether the operation eventually completes.