The SonicPress Times 🎙️

Your Personalized NewsBriefing - Built for ElevenLabs x a16z AI Agents Hackathon 2025

Demo

Home Screen

Headlines and Highlights

Voice Examples

• 🌅 Morning Calm:

sonicpress_news_elevenlabs.mp4

• 🌙 Evening Energetic:

sonicpress_news_mistralai.mp4

• ⚡ Breaking Urgent:

sonicpress_news_andrej_karpathy.mp4

Features

• Personalized News Aggregation: Multi-category news curation with custom topic support
• AI-Powered Summarization: Intelligent content distillation using Mistral AI
• Smart Relevance Filtering: Ensures only topic-relevant articles are included in your briefing
• Dynamic Voice Synthesis: ElevenLabs-powered narration with 3 distinct modes:
  • 🌅 Morning Calm: Warm, relaxed delivery
  • 🌙 Evening Energetic: Dynamic, engaging style
  • ⚡ Breaking Urgent: Authoritative, urgent tone
• Real-time Video Generation: Automated video creation with dynamic image fetching
• Cloud Integration: Google Cloud Storage for asset management
• Modern Web Interface: Streamlit-based UI with NYTimes-inspired styling
• Cloud-Ready: Optimized for Google Cloud Run deployment

System Architecture

The SonicPress system uses an agentic approach with asynchronous tool calls to generate personalized news content:

flowchart TD
    %% User interaction and Streamlit
    User([User]) --> StreamlitUI[Streamlit Web UI]
    StreamlitUI --> UserQuery[User Query Input]
    
    %% Main agent flow
    UserQuery --> InitSystem[Initialize Chat History with System Prompt]
    InitSystem --> StartLoop[Start Loop]
    
    StartLoop --> CallLLM[Call LiteLLM via Proxy API]
    CallLLM --> HasToolCalls{Has Tool Calls?}
    
    HasToolCalls -->|Yes| ExtractToolDetails[Extract Tool Call Details]
    HasToolCalls -->|No| ExtractFinalContents[Extract Final Contents]
    
    ExtractToolDetails --> ToolType{Tool Type}
    
    ToolType -->|get_preferences| GetPreferences[Call get_preferences]
    ToolType -->|fetch_and_summarize| FetchNews[Call fetch_and_summarize via Exa API]
    ToolType -->|generate_news_script| GenScript[Call generate_news_script]
    ToolType -->|text_to_speech| TextToSpeech[Call text_to_speech via ElevenLabs]
    ToolType -->|upload_audio| UploadAudio[Call upload_audio to GCS]
    
    GetPreferences --> StorePreferences[Store Preferences in State]
    FetchNews --> StoreSummaries[Store Summaries in State]
    GenScript --> StoreScript[Store Script in State]
    TextToSpeech --> StoreAudio[Store Audio Path in State]
    UploadAudio --> StoreAudioURL[Store Audio URL in State]
    
    StorePreferences --> UpdateHistory[Update Chat History with Function Result]
    StoreSummaries --> UpdateHistory
    StoreScript --> UpdateHistory
    StoreAudio --> UpdateHistory
    StoreAudioURL --> UpdateHistory
    
    UpdateHistory --> StartLoop
    
    ExtractFinalContents --> CheckForVideo[Check if Audio Exists for Video]
    CheckForVideo --> VideoNeeded{Need Video?}
    
    VideoNeeded -->|Yes| GenerateVideo[Generate Video with MoviePy & FFmpeg]
    VideoNeeded -->|No| DisplayResults[Display Results]
    
    GenerateVideo --> StoreVideo[Store Video Path in State]
    StoreVideo --> DisplayResults
    
    %% Results back to UI
    DisplayResults --> StreamlitUI
    
    %% Cloud Run deployment
    subgraph CloudRun[Google Cloud Run]
        direction TB
        StreamlitUI
        UserQuery
        InitSystem
        StartLoop
        CallLLM
        HasToolCalls
        ExtractToolDetails
        ToolType
        GetPreferences
        FetchNews
        GenScript
        TextToSpeech
        UploadAudio
        StorePreferences
        StoreSummaries
        StoreScript
        StoreAudio
        StoreAudioURL
        UpdateHistory
        ExtractFinalContents
        CheckForVideo
        VideoNeeded
        GenerateVideo
        StoreVideo
        DisplayResults
    end
    
    %% External storage
    UploadAudio --> GCS[(Google Cloud Storage)]
    StoreVideo --> GCS
    GCS --> StreamlitUI
    
    %% Styling
    style CallLLM fill:#f9c6e5,stroke:#d44a7a,stroke-width:2px
    style HasToolCalls fill:#ffe066,stroke:#d4a44a,stroke-width:2px
    style ToolType fill:#ffe066,stroke:#d4a44a,stroke-width:2px
    style VideoNeeded fill:#ffe066,stroke:#d4a44a,stroke-width:2px
    
    style GetPreferences fill:#c6d9f7,stroke:#4a6da7,stroke-width:2px
    style FetchNews fill:#c6d9f7,stroke:#4a6da7,stroke-width:2px
    style GenScript fill:#c6d9f7,stroke:#4a6da7,stroke-width:2px
    style TextToSpeech fill:#c6d9f7,stroke:#4a6da7,stroke-width:2px
    style UploadAudio fill:#c6d9f7,stroke:#4a6da7,stroke-width:2px
    style GenerateVideo fill:#c6d9f7,stroke:#4a6da7,stroke-width:2px
    
    style ExtractToolDetails fill:#f0f0f0,stroke:#333,stroke-width:2px
    style ExtractFinalContents fill:#a8e6cf,stroke:#3a8f5d,stroke-width:2px
    style DisplayResults fill:#a8e6cf,stroke:#3a8f5d,stroke-width:2px
    
    style StreamlitUI fill:#61dafb,stroke:#2d8bba,stroke-width:2px
    style CloudRun fill:#f1f8ff,stroke:#4285f4,stroke-width:2px
    style GCS fill:#c6f7d6,stroke:#4aa76d,stroke-width:2px

Loading

The flowchart illustrates the complete system architecture:

1. User Interaction: Users interact with the Streamlit web interface to input preferences
2. Agent Loop: The NewsAgent runs a loop that checks for tool calls and executes them
3. Decision Points: The system makes decisions based on tool call availability and type
4. Tool Execution: Different tools handle specific tasks (fetching news, generating scripts, etc.)
5. State Management: Results from each tool call are stored in the agent's state
6. External APIs: The system integrates with ElevenLabs, Mistral AI, and Exa for various services
7. Final Processing: After completing all tool calls, the system generates video if audio exists
8. Cloud Infrastructure: Everything runs on Google Cloud Run with media stored in Google Cloud Storage

🛠️ Tech Stack

• Core: Python 3.11+, Poetry
• AI/ML:
  • ElevenLabs API (Voice Synthesis)
  • Mistral AI (Text Processing)
  • LiteLLM for model management
• Frontend: Streamlit
• Storage: Google Cloud Storage
• News Source: Exa News API
• Video Processing: MoviePy, FFmpeg
• Cloud Platform: Google Cloud Run

📋 Prerequisites

• Python 3.11+
• Poetry for dependency management
• FFmpeg installed on system
• API Keys for:
  • ElevenLabs
  • Mistral AI
  • Exa News
• Google Cloud credentials (optional for cloud storage)

🚀 Quick Start

1. Clone the Repository:

git clone https://github.com/yourusername/agentic-news.git
cd agentic-news

2. Install Dependencies:

poetry install

3. Environment Setup:

cp .env.example .env

Edit .env with your credentials:

EXA_API_KEY=your_exa_key
ELEVENLABS_API_KEY=your_elevenlabs_key
MISTRAL_API_KEY=your_mistral_key
GCS_BUCKET_NAME=your_bucket_name (optional)
GOOGLE_CREDS_JSON=path_to_your_creds.json (optional)

4. Run the Application:

poetry run streamlit run streamlit_app.py

☁️ Cloud Run Deployment

Option 1: Using the Deployment Script (Recommended)

The easiest way to deploy SonicPress to Google Cloud Run is using the provided deployment script:

# Make the script executable
chmod +x deploy_to_cloud_run.sh

# Run the deployment script
./deploy_to_cloud_run.sh

The script will:

1. Check for required dependencies (gcloud CLI)
2. Verify you're logged into Google Cloud
3. Create necessary API keys in Secret Manager
4. Create a GCS bucket for media storage
5. Build and push the Docker image
6. Deploy the application to Cloud Run

Option 2: Manual Deployment

If you prefer to deploy manually, follow these steps:

1. Set up API Keys in Secret Manager:

# IMPORTANT: Use echo -n to avoid newline characters in your secrets
echo -n "your_exa_key" | gcloud secrets create sonicpress-exa-key --data-file=-
echo -n "your_elevenlabs_key" | gcloud secrets create sonicpress-elevenlabs-key --data-file=-
echo -n "your_mistral_key" | gcloud secrets create sonicpress-mistral-key --data-file=-
echo -n "your_bucket_name" | gcloud secrets create sonicpress-gcs-bucket --data-file=-
echo -n "your_litellm_proxy_url" | gcloud secrets create sonicpress-litellm-proxy-url --data-file=-

2. Create a GCS Bucket for Media Storage:

gsutil mb -l us-central1 gs://your_bucket_name
gsutil iam ch allUsers:objectViewer gs://your_bucket_name

3. Build and Push the Docker Image:

# Build the image
docker build -t gcr.io/your-project-id/sonicpress:latest .

# Push to Google Container Registry
docker push gcr.io/your-project-id/sonicpress:latest

4. Deploy to Cloud Run:

gcloud run deploy sonicpress \
  --image=gcr.io/your-project-id/sonicpress:latest \
  --platform=managed \
  --region=us-central1 \
  --memory=2Gi \
  --cpu=2 \
  --min-instances=0 \
  --max-instances=10 \
  --timeout=3600 \
  --set-secrets="MISTRAL_API_KEY=sonicpress-mistral-key:latest,EXA_API_KEY=sonicpress-exa-key:latest,ELEVENLABS_API_KEY=sonicpress-elevenlabs-key:latest" \
  --set-env-vars="LITELLM_PROXY_URL=your_litellm_proxy_url,GCS_BUCKET_NAME=your_bucket_name" \
  --allow-unauthenticated

Troubleshooting Cloud Run Deployment

If you encounter issues with your Cloud Run deployment, check the following:

1. API Key Format: Ensure your API keys don't contain newline characters. Use echo -n when creating secrets.

   echo -n "your_api_key" | gcloud secrets versions add sonicpress-exa-key --data-file=-

2. Check Logs: View the service logs to identify specific errors.

   gcloud logging read "resource.type=cloud_run_revision AND resource.labels.service_name=sonicpress" --limit=20

3. Update Environment Variables: If you need to update environment variables or secrets:

   gcloud run services update sonicpress --region=us-central1 --set-secrets=SECRET_NAME=secret-name:latest

4. Verify Configuration: Check the current configuration of your service:

   gcloud run services describe sonicpress --region=us-central1 --format="yaml(spec.template.spec.containers[0].env)"

5. LiteLLM Proxy: If using the LiteLLM Proxy, ensure it's properly deployed and accessible.

🐳 Local Docker Deployment

Build and run with Docker:

# Build the image
docker build -t sonicpress .

# Run the container
docker run -d -p 8080:8080 \
  -e EXA_API_KEY=your_key \
  -e ELEVENLABS_API_KEY=your_key \
  -e MISTRAL_API_KEY=your_key \
  -e GCS_BUCKET_NAME=your_bucket \
  -e GOOGLE_CREDS_JSON=your_creds \
  sonicpress

🎯 Usage

1. Select news categories or enter custom topics
2. Choose your preferred voice mode
3. Adjust content settings (articles per topic, news age)
4. Click "Compile My Headlines"
5. Download the generated video or audio

Voice Settings

Each voice mode has optimized parameters:

VOICE_SETTINGS = {
    "Morning Calm": {
        "stability": 0.71,
        "similarity_boost": 0.85,
        "style": 0.35,
        "speed": 1.0
    },
    "Evening Energetic": {
        "stability": 0.71,
        "similarity_boost": 0.85,
        "style": 0.35,
        "speed": 1.0
    },
    "Breaking Urgent": {
        "stability": 0.71,
        "similarity_boost": 0.85,
        "style": 0.35,
        "speed": 1.0
    }
}

Video Generation

The system automatically creates engaging news videos:

• Dynamic Image Fetching: Retrieves relevant images for each article
• Structured Layout: Organizes content with headlines, images, and summaries
• Scrolling Ticker: Displays headlines in a news-style ticker
• Adaptive Formatting: Adjusts layout based on available content
• Resolution: 1280x720 with H.264 encoding
• Audio: AAC codec, 44.1kHz

LiteLLM Proxy Integration

SonicPress uses a LiteLLM Proxy for improved reliability and performance of LLM API calls. The proxy provides:

• Fallbacks: If one model fails, it automatically retries with another model
• Load balancing: Distributes requests across multiple models or providers
• Unified API: Uses the same API for all LLM providers

Deploying the LiteLLM Proxy to Cloud Run

For optimal performance, we recommend deploying the LiteLLM Proxy to Cloud Run:

# Navigate to the litellm_proxy directory
cd litellm_proxy

# Make the deployment script executable
chmod +x deploy_gcloud.sh

# Set your Mistral API key
export MISTRAL_API_KEY=your_mistral_api_key

# Run the deployment script
./deploy_gcloud.sh

The script will:

1. Build a Docker image for the LiteLLM Proxy
2. Push it to Google Container Registry
3. Deploy it to Cloud Run with the appropriate configuration
4. Return the service URL to use in your application

Using the LiteLLM Proxy Locally

If you prefer to run the LiteLLM Proxy locally for development:

# Set your Mistral API key
export MISTRAL_API_KEY=your_api_key_here

# Navigate to the litellm_proxy directory
cd litellm_proxy

# Build and run the Docker container
docker build -t litellm-proxy .
docker run -d -p 8080:8080 -e MISTRAL_API_KEY=$MISTRAL_API_KEY litellm-proxy

Updating Your Application to Use the Proxy

After deploying the LiteLLM Proxy, update your application to use it:

# Set the LITELLM_PROXY_URL environment variable
export LITELLM_PROXY_URL=https://your-litellm-proxy-url

# Update the application configuration
python litellm_proxy/update_app.py

Troubleshooting LiteLLM Proxy Issues

If you encounter issues with the LiteLLM Proxy:

1. Check Proxy Logs: View the proxy service logs to identify errors.

   gcloud logging read "resource.type=cloud_run_revision AND resource.labels.service_name=litellm-proxy" --limit=20

2. Verify API Keys: Ensure your Mistral API key is correctly set in the proxy service.

   echo -n "your_mistral_api_key" | gcloud secrets versions add litellm-mistral-key --data-file=-
   gcloud run services update litellm-proxy --region=us-central1 --set-env-vars=MISTRAL_API_KEY=your_mistral_api_key

3. Test the Proxy: Send a test request to verify it's working.

   curl -X POST "https://your-litellm-proxy-url/v1/chat/completions" \
     -H "Content-Type: application/json" \
     -d '{
       "model": "mistral-small",
       "messages": [{"role": "user", "content": "Hello, world!"}]
     }'

Troubleshooting "No articles found" Issues

If you encounter "No articles found" errors, try the following solutions:

1. Check API Keys: Ensure your API keys are correctly formatted and don't contain newline characters.

   # Update the EXA_API_KEY in Secret Manager (remove any newline characters)
   echo -n "your_exa_api_key" | gcloud secrets versions add sonicpress-exa-key --data-file=-
   
   # Update the service to use the latest version
   gcloud run services update sonicpress --region=us-central1 --set-secrets=EXA_API_KEY=sonicpress-exa-key:latest

2. Verify Environment Variables: Make sure all required environment variables are set in your service.

   # Check the current configuration
   gcloud run services describe sonicpress --region=us-central1 --format="yaml(spec.template.spec.containers[0].env)"

3. Broaden Search Parameters: Try using more general topics or a broader date range.

   • In the UI, select broader categories like "Tech and Innovation" instead of specific technologies
   • Increase the "News age (days)" slider to search for older articles

4. Check Logs for Specific Errors: Look for error messages in the logs.

   # Check for errors related to the Exa API
   gcloud logging read "resource.type=cloud_run_revision AND resource.labels.service_name=sonicpress AND textPayload:Exa" --limit=20
   
   # Check for errors related to the Mistral API
   gcloud logging read "resource.type=cloud_run_revision AND resource.labels.service_name=sonicpress AND textPayload:Mistral" --limit=20

5. Verify LiteLLM Proxy Connection: Ensure the LiteLLM Proxy is working correctly.

   # Check if the LiteLLM Proxy URL is correctly set
   gcloud run services describe sonicpress --region=us-central1 --format="value(spec.template.spec.containers[0].env[?(@.name=='LITELLM_PROXY_URL')].value)"
   
   # Check the LiteLLM Proxy logs
   gcloud logging read "resource.type=cloud_run_revision AND resource.labels.service_name=litellm-proxy" --limit=20

🔍 Troubleshooting Common Issues

MoviePy/ImageMagick Errors

If you encounter errors like:

MoviePy Error: creation of None failed because of the following error:
convert-im6.q16: no images defined `PNG32:/tmp/tmpXXXXXXX.png' @ error/convert.c/ConvertImageCommand/3229.

This is related to ImageMagick configuration. Here's how to fix it:

1. For Local Development:

   • Ensure ImageMagick is installed: apt-get install imagemagick
   • Modify the ImageMagick policy file:

     sudo nano /etc/ImageMagick-6/policy.xml

   • Update resource limits and permissions:

     <!-- Increase memory limits -->
     <policy domain="resource" name="memory" value="8GiB"/>
     <policy domain="resource" name="map" value="4GiB"/>
     <policy domain="resource" name="disk" value="8GiB"/>
     
     <!-- Allow PDF operations if needed -->
     <policy domain="coder" rights="read|write" pattern="PDF" />

2. For Docker/Cloud Run Deployment:

   • The Dockerfile already includes the necessary configuration
   • If you're still encountering issues, check that the ImageMagick policy is being properly modified during the build process
   • You can verify the policy file inside the container:

     docker exec -it your_container_id cat /etc/ImageMagick-6/policy.xml

🤝 Contributing

1. Fork the repository
2. Create your feature branch (git checkout -b feature/amazing-feature)
3. Commit your changes (git commit -m 'Add amazing feature')
4. Push to the branch (git push origin feature/amazing-feature)
5. Open a Pull Request

Hackathon Context

Built for the ElevenLabs x a16z AI Agents Hackathon 2025, focusing on:

• Voice AI Innovation
• Autonomous Agents
• Real-world Applications

Acknowledgments

• Exa for news search API
• Mistral AI for language processing
• Google Cloud Platform for infrastructure

👨‍💻 Contributors

• Raghav Gupta
• Hrishabh Bhargava