Guide

Phone and voice

Spoken replies on the agent host, and phone calls when the stack is present.

Spoken replies

Spoken replies are generated on the agent host. They are not sent to a cloud text-to-speech service. The default voice is en_GB-cori-medium (British English, public-domain LibriVox).

Voice is turned off on this server.

• A spoken reply is limited to 2,000 characters (the server clamps that setting between 32 and 8,000).
• The speak route allows 10 requests a minute (clamped between 1 and 60).
• The microphone in the browser is treated as cloud speech-to-text. Local-only mode turns it off.
• A local transcription route is present as a stub and is not a finished speech-to-text product.

Phone calls

Hawkeye does not buy phone numbers and does not add a telephony binding. A partner can attach an E.164 number they already control, or provision a bridge. Audio for a call stays on the existing voice-provider session. The Hawkeye websocket accepts text utterances, not binary audio.

The voice-call stack is enabled on this server. See Phone endpoints.

The Receptionist template drafts what to say on a call. It does not place the call by itself. See Templates.