Privacy and data

Heard can run with local components or optional hosted services. The data boundary depends on the speech, narration, and product path you choose.

Part	Local path	Hosted path	What is processed
Speech output	Kokoro	Managed voice, ElevenLabs, or Speechify	Only the text Heard decided to speak
Narration	Neutral templates	Managed narration or your Anthropic key	Compact event context and relevant agent response text
Dictation	Heard Whisper	No hosted service required	Microphone audio stays on your Mac
Session events	Local hooks and local session logs	Optional narration processing after local collection	Events stay local until a selected hosted feature processes relevant text

Managed app

The managed app provides hosted voices and narration without asking you to configure provider keys. Relevant text is sent to the selected hosted service when a cloud voice or narration feature needs it.

Open source engine

The Apache 2.0 engine can use local Kokoro speech and neutral templates without a Heard account. You may also supply your own provider keys directly to the engine.

Coding agent integrations

Claude Code and Codex CLI hooks send events to Heard locally.
The Codex app observer reads local session activity on the same Mac.
The universal wrapper reads output exposed by the wrapped terminal process.
Speech generation and narration happen after Heard decides an event deserves airtime.

Dictation

Heard Whisper performs speech recognition on device. Microphone audio does not need to leave your Mac for push to talk or ambient transcription.

Fully local means using the open source engine with local Kokoro speech and neutral templates. Choosing a cloud voice or hosted narration service sends the minimum relevant text required for that feature.

For legal terms, see the Heard privacy policy and terms.