Who it is forA developer who wants to give an agent a longer instruction by voice from a phone instead of typing it.

The problemTyping a long prompt on a phone keyboard is slow, and the developer switches languages mid-sentence.

The outcomeThe spoken words appear as text in the chat or terminal line for review, and the agent receives them only after the developer sends them.

What you need first

herdr 0.9.0 or newer on the computer that runs the agents
Linux with glibc (x64 or arm64), macOS (Intel or Apple Silicon), or Windows x64 with Git for Windows
curl, Git and tar installed on Linux or macOS; the installer adds Bun 1.4+ and Node 18+ for your user if they are missing
Tailscale or another route to your PC if you want to reach it from a phone

Steps

Install the client with curl -fsSL https://herdrweb.dev/install.sh | sh on the PC that runs herdr.
Open the agent pane in the phone browser and tap the dictation control in the chat.
Speak the message, mixing Korean and English if needed.
Review the transcribed words and correct them before sending.
Send the message to the agent, or queue it for the next turn.

Prompt for your agent

Use the repo devswha/herdr-web-ui. Check whether dictation is configured on this PC by reporting whether an OpenAI key is set or the browser speech recognition is in use. Expected output: a one-line status naming the transcription source. Verify the source by checking the app settings before reporting.

Caveat. Voice input sends audio to the configured provider, or to the browser's speech service, so use it only where that data flow is acceptable.

Use this in your AI agent

Ask your agent

Using the Deduped MCP, get the setup steps for devswha/herdr-web-ui and "Dictate prompts to an agent by voice"

Add Deduped to Claude Code once

claude mcp add --transport http deduped https://dedupedai.com/mcp

Cursor, ChatGPT, others

Third-party repos we don't own. AI-written use cases can be wrong. Disclaimer