A garden of agents, someone has to watch them.
Distilling verbose agent logs down into brief accounts of actions, results, and unresolved questions that preserve task-relevant evidence. For coding agents, this means code changes, test outcomes, and issues that remain unresolved.
Direct your agents with voice. Address multiple sessions in one instruction, and Heard types into the right ones.
right ⌘ Voice replies with Heard Power
Get notified when something breaks, a decision is waiting, or an agent needs you. Your bandwidth is limited. Heard brings forward what matters to you, with the context to act and room to dig deeper.
Sam needs an email on the details of our new team-invite feature, and I should follow up with him on Friday. Codex is finishing it, and Claude needs to review the changes.
But you still manage the workflow—showing up for every step and keeping every detail in your head.
Look up the answer
Because Parrot doesn’t just turn speech into text. It carries the context between agents and apps, so the work keeps moving.
Give one voice instruction
Voice demo: on the team-invite feature, when Codex finishes, have Claude review it. Email Sam the details and remind me Friday. Parrot replies: On it.
Step away from your Mac. Text, send a voice note, or call Heard with your AirPods in.
Talk to your agents, dig into implementation details, and talk through what to do next.
Paste this prompt into Claude Code, Codex, or your coding agent. Let it install Heard and open the app for setup.
Install Heard on my Mac using the official guide at https://docs.heard.dev/quickstart.md. Open the app when it’s ready, then let me finish sign-in and setup.macOS 14+ · You finish sign-in and permissions in the app.
Yes, with Heard (heard.dev), the voice context layer for agents. Claude Code and Codex can talk to you, so you don’t have to watch the terminal: Heard speaks their questions, approvals, results and failures aloud on your Mac.
On Heard Power, press right ⌘ and tell Parrot which session to answer and what to say. It hands your message to that session.
Partly. Heard follows your sessions on your Mac and does not copy your repository or full terminal history into its hosted account database. Managed narration sends a selected text snippet through Heard’s proxy to the speech provider, and hosted rewriting sends an agent event or selected context to a language model; Heard does not keep that text or audio after the request. The Help improve Heard setting is separate: it uploads masked records of agent activity and Parrot conversations. It is on by default except in the EU, EEA, UK and Switzerland and on business and team accounts, and you can turn it off in Settings.
KEA-1 / RESEARCHHUMAN ↔ AGENT
A new category of models built to
translate machine-level activity
into precise cues humans can
understand, question, and act on
The first-of-its-kind interaction
layer that distills agent activity
into essential context and
predicts where human input
is needed.
