Edict for AI Coding

Offline voice dictation for Mac. Tap or hold a key, talk, verbatim text lands at your cursor

Edict is dictation for macOS, built for developers who drive AI coding agents. Tap or hold Right Command, talk, and verbatim text lands at your cursor in any app, including Claude Code and Cursor in a terminal. The speech models run inside the app on the Apple Neural Engine, so dictation works offline from first launch and no audio ever leaves your Mac. Requires macOS 15 or later on an Apple Silicon Mac.

Key features

• Bundled speech models (NVIDIA Parakeet v3) that run on-device. No cloud, no API key, works offline from first launch.

• Verbatim output with no AI rewrite, plus a personal CSV dictionary to fix words it mishears (like "cloud" for "Claude").

• A wrong-window guard: if focus moves while you talk, the text stays on your clipboard instead of pasting into the wrong app.

• Auto-send for agents you allowlist, so a dictated prompt sends itself.

• On-device file transcription with speaker labels, exporting SRT, VTT, or TXT. A 60-minute recording transcribes in around 24 seconds.

• A local MCP server so your coding agent can search transcript history or hand a file over for transcription.

How it works

Tap or hold Right Command and speak. Edict transcribes on the Neural Engine and pastes the text where your cursor is, even in a terminal or over SSH. Drop an audio or video file onto the app to transcribe it on-device.

What's new in 1.0.2

• A smoother first run: onboarding starts with the two permission grants, then shows you the keys.

• The dictation pill fits long app names, and now reports microphone problems (muted, silent, or a reconnected headset) instead of failing quietly.

• System requirements and pricing

• macOS 15 or later, Apple Silicon (M1 or newer).

• 7-day fully functional free trial, no account required.

• One-time license purchase with a 14-day refund. No subscription, no recurring fees.

Keywords: dictation, speech to text, voice typing, offline transcription, on-device AI, privacy, Claude Code, Cursor, coding agents, MCP, macOS, Apple Silicon, SRT, VTT, speaker diarization