Dictate into Cursor, Claude Code, ChatGPT, Cline, Windsurf, Aider — any terminal or editor — because Voicetypr pastes where your cursor is. Then drive transcription programmatically with a scriptable Agent CLI and a local HTTP API. It's also the voice layer for personal AI agents like OpenClaw, NanoClaw, and Hermes — speak to them wherever they live. Offline-first, with local Whisper and Parakeet models.
Two ways to use it
Capability 1 — dictate in
Voicetypr pastes text into whatever field has your cursor. So it works in Cursor, Claude Code, ChatGPT, Cline, Windsurf, Aider — any terminal or editor. There's no plugin to install and no special integration to maintain. Hold the hotkey, describe the change, and the prompt lands where you're typing.
Capability 2 — drive it
A scriptable Agent CLI and a local HTTP API (available while Network Sharing is on) let agents and scripts call transcription directly. Pipe an audio file in, get text out, choose the model and engine per call, and route to a running instance over the network when you need to.
Works in your stack
Because Voicetypr pastes into the focused field, there's no per-tool setup. The same global hotkey works wherever you type.
Personal AI agents
The new wave of self-hosted personal agents runs in the messaging apps and terminal you already use. Voicetypr is the system-wide voice layer: hold one hotkey and talk to your agent in whatever channel it lives in, on Mac or Windows.
The self-hosted personal AI assistant that lives in WhatsApp, Telegram, Slack, and the apps you already use. Dictate to it instead of thumb-typing.
OpenClaw on GitHub →The lightweight, container-isolated personal agent. Talk to it in your messaging channels with one hotkey — the voice step stays on your machine.
NanoClaw on GitHub →Nous Research's memory-keeping agent across Telegram, Discord, Slack, and the terminal. One system-wide voice layer for all of it.
Hermes on GitHub →These agents are self-hosted and privacy-minded — and so is Voicetypr. In local mode your speech is transcribed on your own machine and never sent to us, so the voice layer stays private regardless of how your agent is set up.
Transcription CLI
Run transcription from a shell, a build step, or an agent tool. Add --json for structured output you can pipe anywhere.
$ voicetypr transcribe note.wav --json
{
"text": "Refactor the auth middleware to use the new token store.",
"words": [ /* per-word timing */ ],
"metadata": { "duration": 4.2, "language": "en" },
"model": "parakeet",
"engine": "local"
}
$ voicetypr record --until-silence --json
$ voicetypr transcribe clip.wav \
--model whisper --engine local \
--server 192.168.1.20:8080 --password ••••Built for builders
FAQ
No plugin is needed. Voicetypr pastes transcribed text into whatever input field has your cursor, so it works in Cursor, Claude Code, ChatGPT, Cline, Windsurf, Aider, and any terminal or editor without a special integration.
Yes. A scriptable Agent CLI and a local HTTP API (available while Network Sharing is on) let agents and scripts run transcription programmatically. Use commands like voicetypr transcribe --file <path> and voicetypr record --until-silence, and add --json for structured output.
Not yet. Today you integrate through the Agent CLI and the local HTTP API. Both return machine-readable JSON, so they're straightforward to wire into agent tooling.
Yes. Transcription runs locally by default using Whisper and Parakeet models, so no audio is sent to a cloud service in local mode. Optional cloud STT and optional AI text formatting are opt-in, and AI formatting sends text only, using your own provider and key.
Add --json to a CLI call. For example, voicetypr transcribe note.wav --json returns { text, words, metadata, model, engine }. You can also set --model and --engine per call, and use --server <host:port> with an optional --password to route to a running instance.
Related guides
Narrow to a single tool, see the broader dictation story, or learn how scripts reach a running instance over the network.
single-tool focus
For developers who mainly want to dictate prompts, PR notes, and context straight into the Cursor composer.
broader workflow
The general pay-once dictation story for builders deciding whether voice belongs in their daily setup.
drive it remotely
How the local HTTP API and --server routing let agents and scripts reach a running instance over the network.
3-day free trial. No credit card. Dictate into Cursor and Claude Code, and call transcription from your own scripts.