Skip to content

Latest commit

 

History

History

pipecat-starters

README.md

Pipecat Cloud Starters

This repository contains starter templates for building AI agents using the Pipecat framework and deploying them to Pipecat Cloud.

Available Starters

  • voice - Voice conversation agent with STT, LLM and TTS
  • twilio - Telephony agent that works with Twilio
  • natural_conversation - Text chat agent focused on natural dialogue
  • openai_realtime - Agent using OpenAI's streaming capabilities
  • gemini_multimodal_live - Multimodal agent using Google's Gemini models
  • vision - Computer vision agent that can analyze images

Getting Started

Each starter contains:

  • A functioning bot.py file with a complete agent implementation
  • A Dockerfile for containerization
  • Required dependencies in requirements.txt
  • A targeted README with specific customization options

Prerequisites

Building and Deploying

For detailed instructions on building, deploying, and running your agent, please refer to the Pipecat Cloud documentation.

  1. Choose a starter: Navigate to the starter that best fits your use case

  2. Customize your agent: Modify the bot.py file to change the behavior of your agent

  3. Build the Docker image:

    docker build --platform=linux/arm64 -t my-agent:latest .

  4. Push to a container registry:

    docker tag my-agent:latest your-repository/my-agent:latest
docker push your-repository/my-agent:latest

  5. Deploy to Pipecat Cloud:

    pipecat deploy agent-name your-repository/my-agent:latest --secrets my-secrets

  6. Start a session:

    pipecat agent start agent-name --use-daily

Setting Up API Keys

Different starters require various API keys, which should be added to your Pipecat Cloud secrets.

pipecat secrets set my-secrets \
  OPENAI_API_KEY=sk-... \
  DEEPGRAM_API_KEY=... \
  CARTESIA_API_KEY=...

Documentation

For more information on the Pipecat framework and Pipecat Cloud:

License

This project is licensed under the BSD 2-Clause License - see the LICENSE file for details.