Edict for Mac

Edict, developed by Edict, is an offline voice dictation application for Mac that connects spoken developer intent to coding agents and terminals. The app performs on-device speech-to-text on the Apple Neural Engine and exposes a local MCP server, verbatim transcription tuned for technical vocabulary, push-to-talk hotkey, and high-speed audio/video transcription with SRT export. It targets software engineers, data scientists, and power users who want hands-free interaction with agentic coding tools.

What tasks can you actually use it for?

Edict focuses on translating spoken developer intent into typed input for agentic and terminal workflows. The tool generates verbatim prompts and can paste transcribed lines into active applications to interact with coding assistants. It also converts prerecorded audio and video into transcripts with speaker diarization and SRT/VTT export, so dictation serves both live command entry and batch transcription of meetings or code review recordings.

How accurate are the transcriptions for technical speech?

The app uses NVIDIA Parakeet v3 models for on-device recognition and a verbatim transcription mode tuned for code, identifiers, and command-line jargon rather than editorial rewriting. Users can supply a custom CSV dictionary to correct project-specific tokens. Processing on the Apple Neural Engine produces faster-than-real-time output, though transcription quality depends on microphone clarity, speech pace, and background noise, as with any local speech model.

What file formats and inputs does it handle, and what are the limits?

Edict accepts audio and video files for batch transcription and exports SRT or VTT with optional speaker diarization. For live input it supports a global push-to-talk hotkey (default Right Command) and direct typing into applications. A clear constraint is platform scope: the app requires macOS 15 or later on Apple Silicon, so hardware and OS availability determine whether on-device processing is possible.

Does it fit naturally into developer workflows?

The local MCP server and clipboard-focused design let agentic assistants search recent spoken context, and safety features like Wrong-Window Guard prevent accidental pastes when the active window changes. Auto-send can trigger the Enter key for allowlisted agent terminals, reducing manual keystrokes during prompt entry. The zero-cloud architecture keeps audio and transcripts on the device, matching workflows that disallow sending proprietary code or prompts to external servers.

A practical judgment on who should adopt it

Community feedback describes Edict as generally well-received within developer circles, and the project is produced by a solo creator. Those facts suggest the app suits individual engineers seeking a niche, maintained-by-one utility; teams that require vendor-level support or broad platform coverage should evaluate alternatives that offer organized, multi-person maintenance and institutional backing. Try the tool on a personal project before wider rollout.