Local first. Cloud only when you choose it.

1. Local-first default

Local transcription runs on your device with supported on-device models. Audio and transcripts for local features are not sent to Sawoice servers or cloud AI providers unless you explicitly enable and use an optional cloud feature.

Local models may differ in language coverage, speed, accuracy, memory, storage, and operating-system support.

2. What happens during a cloud request

• The client encodes the requested audio or text in memory and sends it over encrypted transport to the Sawoice API.
• Sawoice authenticates the account, checks entitlement/quota and the current Cloud Processing acknowledgement, then relays the content to the selected provider.
• The provider processes the request under its own Privacy Policy and Terms and returns the transcript or audio response.
• Sawoice returns that response to the requesting client and does not retain the audio, realtime frames, prompt, or transcript body after the response by default.

3. Versioned acknowledgement

A cloud route requires acknowledgement of this version after provider-specific notice. If the acknowledgement is missing, the API returns HTTP 428 with the provider and required version. Local transcription does not require this acknowledgement.

The record shows that the user requested remote processing after notice. It is not described as privacy consent and does not obtain recording permission from a speaker or bystander.

4. Named providers and linked policies

• Nova transcription uses Together AI.
• Scribe batch and realtime transcription use ElevenLabs.
• GPT Transcribe uses OpenAI.
• Grok TTS uses xAI.

For each provider, review the Privacy Policy and Terms links on this page. Retention, training, residency, and acceptable-use rules are set by that provider and may change. Sawoice does not claim zero-retention or identical rules across providers.

5. No zero-retention operating claim

Sawoice does not publicly assert that any cloud provider is configured for zero data retention, zero retention, modified abuse monitoring, or a specific residency region unless that fact is separately confirmed, recorded in the compliance contract, and version-bumped.

The public operating posture is optional cloud relay under each provider's published policies, plus Sawoice's own rule that it does not retain cloud audio or transcript bodies by default.

6. Sawoice metadata

Sawoice retains content-free identifiable request metadata for 13 months: account/device reference, provider, model, route, timestamps, duration, latency, quota/cost calculation, and stable outcome code. Audio, transcript, text, prompt, tokens, raw provider responses, and raw error messages are excluded.

7. Provider-controlled copies

After a cloud request, a provider may retain or process content under its own policy and contract. Sawoice cannot promise deletion of a provider copy it does not control. Use the provider's rights or support channel when needed, and contact support@sawoice.com for help coordinating a request.

8. Sensitive and consequential content

Do not rely on cloud or local AI output without human review. Do not use Sawoice for medical, legal, emergency, financial, or safety-critical decisions. Follow the selected provider's acceptable-use and restricted-content rules, including any limits on sensitive, medical, or financial information.

9. Mobile note exception

The returned transcript is transient unless the user explicitly saves it as a mobile note. An explicitly saved note is stored until hard-deleted or the account closes. Mobile privacy mode blocks automatic transcript-to-note persistence.