OpenTypeless

Privacy-Aware Voice Input: Map Audio, Transcripts, and Polished Text

|By tover0314|12 min read

Privacy in a voice workflow depends on configuration. OpenTypeless handles three different data artifacts—recorded audio, the transcript returned by STT, and polished text sent to or returned by an LLM—and each artifact can take a different route.

Start with the Three Data Artifacts

  • Audio: the captured recording supplied to the configured STT route.
  • Transcript: the text returned by STT before optional polishing.
  • Polished text: the transcript and instructions processed by the configured LLM route.

How Each OpenTypeless Route Changes the Flow

Managed Cloud

With managed Cloud selected, the relevant audio or text is routed through the OpenTypeless service to its configured provider. Check the current Cloud terms and provider policy for retention, logging, region, and account controls before using sensitive material.

BYOK STT and LLM

With BYOK, OpenTypeless sends audio to the STT endpoint and transcript text to the LLM endpoint you configure. BYOK changes credentials and routing control; it does not by itself determine retention or training policy, so verify both providers separately.

Custom Whisper-compatible STT

A Custom Whisper-compatible endpoint is the local or self-hosted STT path. Audio goes to the base URL you enter. If that URL points to a service on the same machine, the intended STT route is local; if it points across a LAN, VPN, or internet connection, audio follows that network route. Check the endpoint and network rather than assuming locality from the label.

Apple Speech

Apple Speech is available as a platform-specific STT path in the supported macOS environment. Audio processing behavior depends on macOS, language availability, permissions, and system settings. Verify the current OS behavior for the device instead of assuming that every Apple Speech request is offline.

Local LLM with Ollama

Ollama is a local LLM provider for text polishing and other LLM work; it is not an STT provider. The transcript and polishing prompt are sent to the configured Ollama base URL. Check that URL and the host network if local-only text processing is required.

Mixed Configurations Are Normal

Local STT with a cloud LLM keeps the audio on the configured local STT route, then sends the transcript and prompt to the cloud LLM. Cloud STT with a local LLM sends audio to the cloud STT provider, then sends the returned transcript to the configured local LLM. Evaluate both halves; choosing one local component does not make the entire workflow local.

Verification Checklist Before Sensitive Dictation

  • Write down the selected STT provider, base URL, and account.
  • Write down the selected LLM provider, base URL, and account.
  • Check whether polishing is enabled and what transcript text it receives.
  • Verify provider retention, logging, training, region, and deletion controls.
  • For local processing, record the Custom Whisper-compatible STT endpoint and the separate Ollama LLM route, then verify both addresses and observe network behavior.
  • Re-check the configuration after app, OS, provider, or policy changes.

Retention, Compliance, and Risk

A routing choice can reduce exposure, but compliance depends on the complete deployment: device access, endpoint operators, contracts, retention settings, logs, backups, and organizational controls. Verify the requirements that apply to your use case; no provider label makes a workflow GDPR or HIPAA compliant by itself.

TIPUseful privacy review asks “where does each artifact go?”—not simply “is this app local?” Check audio, transcript, and polished text independently.

Configuration Examples

  • Managed Cloud STT + managed Cloud LLM: check the service terms for both stages.
  • BYOK STT + BYOK LLM: check both provider accounts and policies.
  • Local or self-hosted route: a Custom Whisper-compatible endpoint handles STT, while Ollama handles the separate local LLM polishing stage; verify both routes and any model-download or update traffic.
  • Apple Speech + BYOK or local LLM: verify macOS behavior and the separate LLM route.

Open Settings before recording sensitive material, verify the STT and LLM selections, and run a non-sensitive test. Revisit this map whenever the configuration or provider policy changes.