Data handling

Minimize the content you send and choose persistence intentionally.

By default, Enterprise API telemetry is metadata-only. Standard telemetry does not include request or response content, cleaned context, or source audio.

Set store to true only when your product intentionally opts into request, response, and audio persistence:

• For batch dictation, send store=true as a multipart form field.
• For realtime streaming, send "store": true in the initial session.start message.

Stored realtime sessions include the streamed audio, raw and final transcripts, context, dictionary, and extra instructions. Realtime PCM audio is retained as a mono 16 kHz WAV file. Your enterprise agreement may define additional handling or retention requirements.

Submitting transcription feedback does not retroactively enable full dictation storage. The feedback signal and identifiers are stored, plus only the optional text or audio that the client explicitly attaches.

Send the minimum useful context

• Prefer cursor-adjacent text over an entire document.
• Include the current app, window title, or selection only when it helps the result.
• Put names and terminology in dictionary.
• Put tone and output behavior in extra_instructions.
• Remove passwords, API keys, authentication codes, payment details, and unrelated private content.

Correlate without embedding private data

Use transcription_id, platform, and app_version for diagnostics. Do not put user IDs or internal identifiers into context unless the model genuinely needs them to produce the requested output.

Before production, document your own consent, retention, deletion, and access-control behavior for end users. Contact your Monologue representative when contractual requirements differ from the defaults described here.