SpeakEasy

SpeakEasy is the voice interface for the Codex task you are already using. It combines local speech recognition, exact-task routing, and controllable narration without creating a second conversation.

Choose your path

SpeakEasy for Codex

Use the native macOS app when you want to:

• dictate into an exact Codex Desktop task from the Mac, browser, or iPad;
• keep the complete transcript and tool activity in Codex;
• hear the real response through a native player with queue, replay, speed, volume, and provider controls;
• map up to nine persistent voice lanes without relying on window focus.

Start with Listening mode or the Codex installation page.

Requirements: an Apple silicon Mac running macOS 14 or newer and the current Codex Desktop app. The packaged app is self-contained; Bun, Node, Xcode, and a source checkout are not required.

TypeScript library and CLI

Use @arach/speakeasy when an application, script, or agent hook needs provider-independent text-to-speech.

bun add @arach/speakeasy

import { say } from '@arach/speakeasy';

await say('The build is ready.');
await say('Use the configured OpenAI voice.', 'openai');

Library requirements: Node.js 22.12 or newer, or Bun 1.0 or newer. Native system playback is macOS-only; cloud providers require their own API keys.

Product loop

1. Choose the real Codex task that owns the work.
2. SpeakEasy validates and locks that exact task ID.
3. Hold to speak. Audio is transcribed locally on the Mac.
4. SpeakEasy submits the final text exactly once to the locked task.
5. The full response stays in Codex and is narrated through SpeakEasy.

If task identity, ownership, or turn correlation cannot be proven, SpeakEasy fails closed instead of selecting another task.

Documentation

• Quickstart — install the app or library
• Listening mode — exact-task voice turns and trust boundary
• CLI reference — speech and cache commands
• SDK guide — TypeScript API
• Configuration — providers and global settings
• Providers — System, OpenAI, ElevenLabs, Groq, and Gemini
• Custom providers — add a TTS adapter
• Cache — local SQLite cache behavior
• Troubleshooting — diagnostics and common failures

Release status

The signed 0.3.0 build is the first complete dual-mode release. It includes the paired-HTTPS browser/PWA Deck alongside exact-task dictation and narration on the Mac. The native iPad shell remains a developer preview.

See the GTM control plan for the launch sequence and the second-Mac checklist for the proof gate.

Support

Run speakeasy --doctor for library and provider diagnostics. For the native app, open SpeakEasy → Settings → Deck and use Open log when a readiness check fails.