Tethro

Voice-first AI proxy

A voice-first personal agent that remembers people and commitments, drafts messages, and asks before it acts on your behalf.

Purpose

Tethro is a personal chief of staff for relationship work. Spoken conversation becomes structured records of people, memories, promises, and follow-ups without requiring a form after every meeting.

When a message or other outgoing action is requested, Tethro prepares a draft and stops for confirmation.

Voice to record

Voice Dump parses unstructured speech into person, memory, and promise records. A realtime voice path runs over a WebSocket with interruption handling, while a lower-cost streaming path handles routine turns.

Input: Unstructured speech.
Record: People, memories, promises, and due dates.

Memory and confirmation

Person lookup moves from exact match to vector, phonetic, and fuzzy matching before asking for clarification. Dates are parsed by a rule-based engine rather than generated by a language model.

Every write passes through a six-state Confirmation Gate. The action is held, shown to the user, and executed only after explicit approval.

Recall path: Exact → vector → phonetic → fuzzy → ask.
Outgoing actions: No write executes without explicit approval.