Skip to Content
DocsFeaturesVoice — Talk to AIDEN

Voice — Talk to AIDEN

Talk to AIDEN is the voice orchestrator built into the desktop app.

Toggle voice mode, hold Space, and say what you want. Quick requests happen instantly inside the app. Heavier work — coding, research, file edits — is delegated to a background agent, and AIDEN reads the result back aloud when it finishes.

It is push-to-talk, not always-on: the microphone is only live while you hold the key.

Setting up voice

Voice needs a one-time setup before the first session.

  1. Open Settings → Voice.
  2. Add an OpenAI API key from platform.openai.com/api-keys .
  3. Grant microphone access when the app asks for it.

Use a platform.openai.com key with billing enabled. A ChatGPT or Codex subscription cannot reach the Realtime voice service — voice needs an actual API key.

Your key is stored locally on your machine and is only used to open voice sessions. It is never displayed again in full — the settings page shows a masked version, with Replace and Remove buttons if you ever want to change it.

While you are in settings, you can also personalize the experience:

  • Voice engine — choose the realtime model (GPT Realtime 2 is the default; a Mini option is available if you prefer lighter and faster).
  • Voice — pick from ten voices, including marin, cedar, and alloy.
  • Delegation agent — choose which worker handles delegated tasks: Auto (best available), Codex, Claude, or the OpenAI API, each with its own model options. Auto is a good default — it uses the CLIs you already have installed.

Starting a session

There are two ways in:

  • Press Cmd/Ctrl + Shift + V anywhere in the app.
  • Click Talk to AIDEN in the top bar.

Then the interaction model is simple:

  • Hold Space to talk. Release to send. (Space is ignored while you are typing in a text field, so it never fights with normal typing.)
  • Hold Space mid-reply to barge in — AIDEN stops talking and listens to your new turn.
  • Press Esc to hard-stop a response or exit voice mode.

While a session is active, the top bar shows a live mic button with a waveform, plus a status label that tells you exactly what is happening: Listening…, Thinking…, Speaking…, or the tool currently running. A soft glow around the window signals that voice mode is on.

Everything you say — and everything AIDEN says back — streams as live captions, and you can open the conversation panel at any time to see the full transcript of the current session.

What you can do by voice

Instant in-app actions

Lightweight requests run the moment you ask:

  • navigate the app — open a project, jump to a page
  • open, split, and arrange panels; open a terminal
  • manage the board — create, rename, move, schedule, or delete stories and epics
  • quick reads — search stories, check a task’s status, list your projects

Delegate real work

Anything heavier is handed to a background agent: implementing a change, researching a question, editing files, running an automation. You can keep working — or step away entirely — while it runs.

A compact HUD in the top bar shows the running agent and narrates its progress step by step: queued, running, then completed or failed. When the task finishes, AIDEN speaks a summary of the result aloud.

Approvals for risky actions

Destructive or outward-facing actions — deleting, sending, deploying, calendar or account changes — never run on voice alone. They are automatically flagged and paused, and Approve / Decline buttons appear in the HUD. Nothing happens until you explicitly say yes.

Requirements

  • macOS 12 or later
  • An OpenAI API key with billing enabled (for the realtime voice layer)
  • A microphone and an internet connection
  • For delegated coding work: at least one of Claude Code or Codex CLI installed — the same setup as the rest of AIDEN

Delegated work runs on your machine through your own CLIs and keys. Your code stays local.

Honest limits

  • Push-to-talk only. There is no wake word and no always-on listening.
  • One delegated agent at a time. The board can run many agents in parallel; the voice delegation slot is single-agent for now.
  • Sessions are in-memory. Voice transcripts are not saved across app restarts.

Tips

  • Treat voice as a second input mode, not a replacement. It shines when your hands are busy, when you want to kick off a task quickly, or when you are thinking out loud about what to build next.
  • Delegated stories still land on the board like any other work — you review every diff the same way.
  • If a reply is going the wrong direction, barge in early. Holding Space is faster than waiting for the answer to finish.
AI ChatCustom Agents