Herd Your
Agents.
Walk around with smart glasses while your AI coding agents build, test, and deploy. VoxHerd is the voice remote control for Claude Code, Codex, and Gemini CLI.
Refactored pricing engine to ISO 8601 dates.
Adding Zod validation schemas for API routes.
src/schemas/api.ts
Set up project structure and initial routes.
Features
Everything you need.
Voice-first control
Speak naturally to your agents. VoxHerd uses on-device speech recognition to parse your intent and route commands to the right project -- no typing required.
Multi-agent orchestration
Run Claude Code, Codex, and Gemini CLI sessions simultaneously across different projects. Each agent works independently while VoxHerd keeps you in the loop on all of them.
Smart glasses ready
Walk around the house, go for a coffee, take the dog out. Your agents report in through Meta Ray-Ban smart glasses via Bluetooth audio -- just listen and respond.
Intelligent routing
Fuzzy project matching, context-aware routing, and compound utterance parsing. Say 'Switch to Homeschool, now add integration tests' and VoxHerd handles both intents.
Zero-config integration
VoxHerd hooks into each agent's lifecycle events. When a session starts, completes, or needs permission, the bridge server is notified instantly. Haiku generates concise summaries that are announced via TTS. Works with Claude Code hooks natively, and Codex and Gemini CLI via tmux integration.
"hooks": {
"stop": [{
"command": "~/.voxherd/hooks/on-stop.sh"
}],
"session_start": [{
"command": "~/.voxherd/hooks/on-session-start.sh"
}]
}
}
Architecture
How it works.
Three components, one voice-driven workflow. No cloud services, no accounts, no auth. Everything runs on your local network.
Agent Lifecycle Hooks
Mac / Linux
Bash scripts and tmux watchers that fire on session start, stop, and permission requests. They POST events to the bridge server with session context and Haiku-generated summaries. Native hooks for Claude Code, tmux integration for Codex and Gemini CLI.
Bridge Server
FastAPI on port 7777
The central nervous system. Receives hook events, manages session state, dispatches voice commands to Claude Code, Codex, or Gemini CLI, and streams everything over WebSocket.
iOS App
iPhone / Smart Glasses
A voice remote control. Captures speech, parses intent, sends commands over WebSocket, and announces results via TTS. Never touches code or files directly.
AI Agents
Your fleet, your voice.
VoxHerd orchestrates any number of Claude Code, Codex, and Gemini CLI instances across your projects. Manage them all with natural conversation.
Works with any terminal-based AI coding agent via hooks or tmux integration.
Conversational Flow
Refactored pricing engine to use ISO 8601 dates. All 47 tests passing.
Great, now add the currency conversion endpoint.
Wants permission to run: npm install zod
Approve it.
Switch to Homeschool. Add integration tests for the auth module.
Working on it...
Get Started
Up and running in minutes.
VoxHerd runs on your Mac or Linux machine. No cloud, no accounts. Just install and start herding.
Install hooks
Deploys hook scripts to ~/.voxherd/hooks/ and patches your agent settings. Supports Claude Code hooks natively, plus tmux-based integration for Codex and Gemini CLI.
Start the bridge
Starts the FastAPI bridge server on port 7777 with text-to-speech announcements.
Connect your phone
Point the iOS app at your bridge server. Start talking to your agents.