🎬 EchoBot

The ultimate framework for AI Agents to go live on YouTube

Microservices-based platform that enables AI personas to autonomously host dynamic live streams — from content generation to OBS studio controls

🎥 See it in Action

Check out a working example of EchoBot in action:

• XyLumira YouTube Channel - A live streaming agent powered by EchoBot

✨ Features

• Microservices Architecture: A scalable and maintainable architecture with dedicated services for each core functionality.
• Dynamic Scheduling: Control your broadcast in real-time by editing a schedule.json file.
• OBS Automation: Seamlessly switch scenes, manage media, and control your stream via OBS WebSockets.
• Autonomous Agent Content: The AI Agent automatically generates news reports, voiceovers, and music to keep the stream fresh and engaging.
• Unified Media Management: A centralized media directory simplifies content management and ensures consistent access across all services.
• Live Agent Interaction: The AI Agent engages with your audience in real-time via YouTube chat, answering questions and reacting to comments.

🏗️ Architecture

Service	Description
API Service	Exposes an API for managing the streaming agent and its services
Chat YouTube Service	Manages YouTube chat interaction and AI-powered responses
Music Service	Handles music generation, downloads, and playback
News Service	Aggregates news and generates scripts for the AI persona
OBS Stream Service	Controls OBS and manages the live stream
Event Notifier Service	Receives events from other services and forwards them to configured webhook URLs

📌 Status

EchoBot is under active development. Community contributions are welcome via pull requests.

---

📋 Prerequisites

Install the following tools before proceeding:

Tool	Purpose
Python 3.13+	Runtime
ffmpeg	Audio/video duration detection
tmux	Service management (required by start_services.py)
OBS Studio	Stream control (has built-in WebSocket server)
Docker	Optional, for containerized deployment

Install via Homebrew (macOS/Linux):

brew install ffmpeg tmux

---

🚀 Quickstart

1. Install uv Package Manager

If you don't have uv installed, install it with the following command:

curl -LsSf https://astral.sh/uv/install.sh | sh

2. Clone and Install Dependencies

git clone <repository-url>
cd echobot
uv sync

3. Create Media Directories

mkdir -p app/media/{videos,voice/generated_audio,music,news,state,memory}

4. Create Environment File

cp .example.env .env

Edit .env with minimum required values:

# OBS Connection
OBS_HOST=localhost
OBS_PORT=4455
OBS_PASSWORD=your_obs_password

# Media root (absolute path to your media folder)
MEDIA_HOST_DIR=/path/to/echobot/app/media

5. Enable OBS WebSocket

1. Open OBS Studio
2. Menu bar (top of screen on macOS) → Tools → WebSocket Server Settings
3. Check "Enable WebSocket server"
4. Set a password and note the port (default: 4455)
5. Click OK

6. Prepare Your Media 🎞️

All media files are stored in app/media/:

app/media/
├── videos/              # Your video files (.mp4, .mov, .webm)
├── music/               # Background music and soundtracks (.mp3, .wav)
└── voice/generated_audio/  # Generated voiceovers (auto-created by news service)

Videos: Place all video files in app/media/videos/

• Use for scene backgrounds (looping animations, DJ visuals, etc.)
• Supported formats: .mp4, .mov, .webm
• Tip: Videos without audio track work best (audio is managed separately)

Music: Place audio files in app/media/music/

• Use for background music that plays continuously
• Supported formats: .mp3, .wav

7. Configure Your Scenes 📅

EchoBot uses two configuration files to control what plays and when:

File	Purpose
config/schedule.json	Scene Registry — defines available scenes with their video/audio files
services/obs_stream_service/core/playlist.json	Execution Sequence — defines which scenes play and for how long

Step 1: Edit config/schedule.json

Define all available scenes and match them to your media files:

{
  "background_music": {
    "enabled": true,
    "file_path": "music/song.mp3",      // Your music file from step 6
    "loop": true,
    "volume_normal": 0.3,
    "volume_ducked": 0.1
  },
  "current_scene": {
    "scene_name": "Scene-Music",
    "video_path": "videos/visual.mp4",
    "has_audio": false
  },
  "_available_scenes": {
    "music": {                           // Scene identifier (use this in playlist.json)
      "scene_name": "Scene-Music",       // OBS scene name (will be auto-created)
      "video_path": "videos/visual.mp4", // Your video file from step 6
      "video_source_name": "MusicVideo", // [Required] Unique identifier
      "has_audio": false,                // [Required] false = music stays at 30%
      "loop_video": true                 // [Required] true = loops forever
    }
    // Add more scenes here as needed
  }
}

Step 2: Edit services/obs_stream_service/core/playlist.json

Define the execution sequence:

{
    "variables": {
        "music_duration": 300           // 5 minutes in seconds
    },
    "playlist": [
        {
            "scene_name": "music",      // References "music" from schedule.json above
            "duration": "$music_duration"
        }
        // Add more scenes here to create a sequence
    ]
}

How it works:

• RadioFlow reads the playlist array sequentially
• For each scene, it looks up the definition in schedule.json
• After the duration expires, it moves to the next scene
• When the playlist ends, it loops back to the beginning
• The system runs fully autonomously — no manual intervention required

Step 3: Auto-Create OBS Scenes

Run this script to create all scenes in OBS automatically:

uv run python -m services.obs_stream_service.scripts.init_obs_scenes

This reads your schedule.json and creates the OBS scenes with video/audio sources.

---

9. Running EchoBot ▶️

Launch the OBS Service

uv run python start_services.py --launch obs

This will:

• Connect to OBS via WebSocket
• Loop through the playlist automatically
• Switch scenes, manage audio ducking, and cycle media sources

How Scene Switching Works

1. RadioFlow reads playlist.json and executes scenes sequentially
2. For each scene, it looks up the definition in schedule.json
3. News/Music services update schedule.json with generated content paths
4. When a scene with duration: null is reached, it waits for audio to complete
5. The playlist loops forever

No manual intervention required — the system is fully autonomous once launched.

Managing Services

# Launch specific services
python start_services.py --launch obs

# Stop all services
python start_services.py --stop

# Stop specific service
python start_services.py --stop obs

# Restart a service
python start_services.py --launch obs --force

Available services: youtube, obs, music, news, api, event_notifier

---

📁 Media Directory Structure

All media paths in schedule.json are relative to MEDIA_HOST_DIR:

app/media/
├── videos/              # Video files for scenes
├── voice/
│   └── generated_audio/ # Generated voiceovers
├── music/               # Background music
├── news/                # Generated news scripts
├── state/               # Service state files
└── memory/              # Agent memory

⚙️ Optional: Advanced Features

The following features require additional API keys and configuration. See .example.env for all available options.

📰 News Generation Service

Generates AI news reports with voiceover. Requires:

• ElevenLabs API key (voice generation)
• LLM provider API key (script generation)

🎵 Music Generation Service

Generates original music. Requires:

• Suno API key or SoundCloud credentials

💬 YouTube Chat Service

Enables live chat interaction. Requires:

• YouTube API credentials
• OAuth configuration

🔔 Event Notifier Service

Forwards events to external webhooks for integration with your website or other services. Configure EVENT_WEBHOOK_URLS in .env.