Free · Source-Available · No API Keys

Hands-free voice
for your coding agent.

Talk to Claude Code, Codex, Antigravity, or Ollama directly from your phone. It does the work and replies in chat and aloud — like a real phone call with your developer companion.

Quickstart View Source
Powerful Features

A seamless bridge between you and your code.

A tiny Node bridge drives the CLI on your machine and streams everything to your phone. Zero latency, total privacy.

🎙️

Type or Speak

Use the chat interface to type, or tap the mic to speak. The agent's replies are formatted beautifully and read aloud sentence-by-sentence.

🤖

Multi-Agent Ready

Switch seamlessly between Claude Code, Codex, Antigravity, and Ollama (fully local). Each has adjustable autonomy modes.

🗂️

Parallel Sessions

Run "Claude on Repo A" and "Codex on Repo B" simultaneously. Jump between ongoing persistent sessions instantly.

📞

Talking Mode

Enter a continuous voice loop. The orb pulsates as you speak, the agent executes, and reads the reply back to you autonomously.

🔒

Private by Default

Everything runs over your private Tailscale network via HTTPS. Your codebase never touches a third-party voice cloud.

⚡

Local-First Option

Uses built-in Browser Web Speech API for free STT/TTS. Or, configure it to use your own local Whisper instance for max accuracy.

Architecture

Phone in, agent out.

Your phone handles the voice layer, while the heavy lifting (the CLI) stays on your Mac or PC.

[ iPhone Safari ] [ Your Mac ] mic ─Web Speech STT─▶ text ──https/Tailscale──▶ voicebridge ──spawn──▶ claude / codex speaker ◀─speechSynthesis─ reply ◀────────────── reply ◀─────────── (agent CLI)
Get Started

Up and running in 4 lines.

You need Node ≥ 18, an authenticated agent CLI, and Tailscale installed.

git clone https://github.com/berkayturanci/speak-with-claude-code voicebridge
cd voicebridge && npm install

# Optional: Set up your environment (PROJECT_DIR, ACCESS_TOKEN)
cp .env.example .env

# Start the server (prints a QR code)
npm start

# In a new terminal tab: Expose securely to your phone
tailscale serve --bg --https=443 localhost:8787

💡 Tip: Scan the QR code with your phone. Ensure Tailscale is active on both devices!

FAQ

Common Questions

Is the voice feature actually free?
Yes. Speech-to-text and text-to-speech run natively in your phone's browser using the Web Speech API. This means zero dependencies on paid services like ElevenLabs. (You only pay for your actual LLM agent usage, like your Anthropic Claude plan).
Does my code leave my machine?
No, VoiceBridge acts only as a local proxy. Your agent (Claude, Codex, Antigravity) runs locally on your machine. Voice recording is transcribed on-device (via Apple/Google's dictation engines) and sent securely over your private Tailscale network.
Why do I need Tailscale and HTTPS?
Modern browsers (Safari, Chrome) strictly enforce security and completely disable microphone access on non-secure origins (like standard `http://192.168.x.x`). Tailscale Serve provides a real, trusted HTTPS certificate over your private network, instantly unlocking mic access.
Can I use Whisper instead?
Absolutely. If you want maximum accuracy (especially for coding syntax and punctuation), you can configure VoiceBridge to use your own local Whisper command-line tool via the `STT_MODE` and `STT_CMD` environment variables.