bodhi_realtime_agent

<p align="center">
  <img src="docs/images/icon_graphic_only.png" alt="Bodhi Agent" width="120" />
</p>

# Bodhi: High-Performance, Multi-Agent realtime Voice Stack

[![npm version](https://img.shields.io/npm/v/bodhi-realtime-agent.svg)](https://www.npmjs.com/package/bodhi-realtime-agent)
[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)

**Conversational voice agents with non-blocking background execution.** 
(No other framework does this out of the box.)

---

Most voice agent frameworks block conversation while tools execute. User says "generate me a video" and the agent goes silent for two minutes. Bodhi splits the work: **main agents** handle the live conversation (Gemini Live API by default, OpenAI Realtime also supported), while **background subagents** (Vercel AI SDK) run long tasks in parallel. When a task finishes, the agent naturally announces it.

```
User: "Make me a video of a sunset AND search for weather in Tokyo"

Main Agent (Gemini/OpenAI Live — realtime voice):
  "I'm generating your video and looking up the weather..."
  │
  ├─ Subagent 1: Video generation (2 min, Veo API)
  ├─ Subagent 2: Web search (3 sec, Gemini)
  │
  "What else can I help with?"      ← keeps talking
  │
  [3s]  "The weather in Tokyo is..."  ← search result arrives
  [2m]  "Your video is ready!"        ← video arrives
```

### Claude Code Demo — Voice-Driven Coding Assistant

<p align="center">
  <a href="https://youtube.com/watch?v=n3kNK-4expo">
    <img src="https://img.youtube.com/vi/n3kNK-4expo/maxresdefault.jpg" alt="Watch the Claude Code demo" width="700" />
  </a>
  <br />
  <em>Click to watch the demo</em>
</p>

Speak coding requests naturally — Claude Code reads, edits, creates files, runs commands, and searches your codebase in the background while you keep talking. See [examples/claude_code](examples/claude_code/) for setup.

### OpenClaw Integration — Voice-Driven AI Agent

<p align="center">
  <a href="htt