Skip to content

nickzsche21/VocalClaw_11Labs

Repository files navigation

🎙️ VocalClaw

Multi-Agent Voice AI — Built for ElevenLabs Agentic Summer Buildathon

Live Demo


What is VocalClaw?

VocalClaw is a multi-agent voice AI system where four specialized AI agents — each with a distinct ElevenLabs voice — collaborate to answer your questions.

You speak. Agents think. You hear the answer — automatically, in the right expert's voice.

The philosophy: LLM inference should cost $0. ElevenLabs should handle everything voice-related. That's the stack.

  • ElevenLabs — Voice I/O (Scribe STT + TTS for all 4 agents)
  • Groq — Free-tier LLM inference (inspired by AirClaw's zero-cost local inference philosophy)
  • No OpenAI. No per-token costs. No paywalls.

VocalClaw is architecturally inspired by AirClaw — the belief that AI inference should be free and open. AirClaw runs locally; VocalClaw extends that philosophy to the cloud using Groq's free API.


🎯 Two Modes

Solo Mode

You ask → Aria detects intent → routes to the right specialist → that agent responds in their unique voice

Aria even speaks the handoff before passing to the specialist. That's agentic behaviour.

⚡ Council Mode — The Killer Feature

You ask ONE question → all 4 agents respond in sequence, each in their own voice → Aria synthesizes a final answer

You literally hear a 4-voice AI debate. No other voice AI does this.


The Agents

Agent Voice Speciality
🎯 Aria Rachel Orchestrator — routes, synthesizes, opens council
💻 Rex Domi Code — programming, debugging, architecture
⚖️ Lex Bella Legal — Indian law, IPC, CrPC, case precedents
🔬 Max Antoni Research — science, history, economics, facts

Architecture

🎤 You speak
     │
     ▼
ElevenLabs Scribe STT
     │
     ▼
┌─────────────────────────┐
│  ARIA — Orchestrator    │
│  Intent detection       │
│  Routes + announces     │
└──────┬──────────────────┘
       │ speaks handoff aloud
  ┌────┼────────┐
  ▼    ▼        ▼
 REX  LEX      MAX
(Code)(Legal)(Research)
       │
       ▼
ElevenLabs TTS — Agent's unique voice
       │
       ▼
🔊 Autoplays instantly

Features

  • 🎤 Mic input — record your question directly in browser
  • 🔊 ElevenLabs Scribe STT — transcribes your voice instantly
  • 🤖 Real AI via Groq (LLaMA 3) — not templates
  • 🗣️ Aria speaks handoffs — announces routing in her own voice
  • Council Mode — 4 agents, 4 voices, 1 question, 1 synthesized answer
  • 🔁 Autoplay — audio fires automatically, no click needed
  • 💬 Session history — full conversation thread

Live Demo

👉 vocalclaw11labs-6sqxrv2exz5ce62sn3n7cy.streamlit.app

Try these:

  • "Explain IPC Section 302" → Lex responds in Bella's voice
  • "Write a Python web scraper" → Rex responds in Domi's voice
  • Council Mode + "Should I raise a seed round or bootstrap?" → hear all 4 voices debate it

Run Locally

git clone https://github.com/nickzsche21/VocalClaw_11Labs
cd VocalClaw_11Labs
pip install streamlit requests
streamlit run app.py

Why This Wins

Most voice AI = one voice, one brain. Generic.

VocalClaw = 4 specialized agents, 4 distinct voices, free inference, spoken handoffs, and a Council Mode where agents literally debate your question out loud.

Lex is a prototype of the world's first voice-native Indian legal AI — built on real domain depth from JurixAI (backed by NSRCEL IIM Bangalore + Perplexity AI Fellowship).


Built By

Nick — Founder, JurixAI

Built in one night for the ElevenLabs Agentic Summer Buildathon #1


License

MIT

About

Run OpenClaw AI agent with zero API cost, local LLM via AirLLM

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors