Heard Parrot

Heard Parrot turns the narrator into someone you can ask. Hold the talk key, address your narrator by name, and ask about the repo you are working in. Heard runs a read-only dig through the codebase and answers out loud.

The wake word is your narrator’s name: “Jarvis” if you use the default voice, or “Aria”, “Friday”, “Atlas” if you chose another. “Hey Parrot” always works, whichever voice is speaking.

Things you can say

"Jarvis, walk me through this codebase."
"Aria, how's the backend doing?"
"Hey Parrot, where does sign-in happen?"
"Jarvis, catch me up."

How it works

Hold the talk key and say the wake word

Hold Right Command as usual and open with your narrator’s name, or “Hey Parrot”. The moment Heard hears the wake word, the pill switches from dictation to Parrot, so you can see the question is going to your narrator instead of your cursor.

Heard digs, then answers

Parrot explores the active repository read-only. It never edits files and never runs your coding agent. Depending on your settings, it can speak brief progress while it digs.

Follow up naturally

After an answer, the conversation stays open for a moment: ask follow-ups without repeating the wake word. Tap the talk key on its own to end the conversation and return the key to dictation.

Catch me up

“Catch me up” gives a spoken recap across your active sessions: what each agent did, what landed, and what is waiting on you. Useful after stepping away, or as the closing beat of a long parallel run.

Read-only, non-intrusive by design

Heard Parrot observes; it does not act on your work.

It never edits files, never runs your coding agent, and never executes commands in your repositories.
It never types into an existing session: a Parrot question is routed away from your cursor, so nothing lands in a terminal by accident.
Its digs through the codebase are read-only, and they run in a separate throwaway context that your agents never see.

The one category of action Parrot can take is opening things: a fresh terminal seat, a Finder window, or bringing an app’s window forward. Opening is non-destructive and visible the moment it happens, and you can turn it off entirely, making Parrot answer-only.

Open things by voice

Parrot includes a small set of safe open commands:

Safe open commands

"Jarvis, open claude in checkout-service."
"Hey Parrot, open codex in my-app."
"Jarvis, open checkout-service in Finder."

Projects in your projects folder are addressable by name, and a .heard.yaml label works here too. You can disable open commands entirely in Settings.

Settings

Setting	What it does
Answer depth	Brief, Balanced, or Deep: how much codebase context Parrot speaks in its first answer.
Progress updates	Whether Parrot narrates short waypoints while it digs.
Disable safe open commands	Turns off the open-by-voice verbs, leaving Parrot answer-only.

Parrot is part of Heard Power and rides the same on-device speech recognition as dictation: hold the key, speak, release. Questions about the codebase stay read-only by design.