Skip to content

Latest commit

 

History

History
184 lines (144 loc) · 6.02 KB

README.md

File metadata and controls

184 lines (144 loc) · 6.02 KB

AWS + Deepgram Voice AI Hackathon: Pipecat Quickstart Project

This is the quickest way to get started building a Pipecat voice AI agent for the AWS + Deepgram Voice AI Hackathon.

It was built with the Pipecat CLI, and customized to make it as fast as possible to get a working bot.

Dependencies

  • Python 3.10+
  • uv
  • docker (to deploy to Pipecat Cloud)

Run your bot locally

git clone git@github.com:pipecat-ai/aws-deepgram-sa-hackathon.git
cd aws-deepgram-sa-hackathon

Server

In your first terminal window:

cd server
cp env.example .env # and fill in the values
uv sync
uv run bot.py --transport daily

You should see something like:

INFO     | pipecat:<module>:14 - ᓚᘏᗢ Pipecat 0.0.102 (Python 3.12.0 (main, Oct  2 2023, 20:56:14) [Clang 16.0.3 ]) ᓚᘏᗢ

🚀 Bot ready!
   → Open http://localhost:7860 in your browser to start a session

INFO:     Started server process [91430]
INFO:     Waiting for application startup.
INFO:     Application startup complete.
INFO:     Uvicorn running on http://localhost:7860 (Press CTRL+C to quit)

Client

Then, in another terminal window:

cd ../client
cp env.example .env.local # and fill in the values
npm i
npm run dev

You should see something like:

  VITE v7.3.1  ready in 206 ms

  ➜  Local:   http://localhost:5173/
  ➜  Network: http://192.168.0.16:5173/
  ➜  Network: http://100.115.25.125:5173/
  ➜  press h + enter to show help

Now visit http://localhost:5173 in your browser and click Connect to start talking to your bot!

Architecture Diagram

graph TD
    subgraph Users
        A["👤 <b>End-user</b>"] --> B["React Web Client<br/><i>Vite + RTVI</i>"]
    end

    subgraph Transport["Daily Server - WebRTC Transport"]
        C[Audio In]
        D[Audio Out]
    end

    subgraph Pipeline["Conversational Voice Agent - Orchestrated by Pipecat"]
        E["Krisp Noise Cancellation<br/><i>Cloud only</i>"]
        F["Silero VAD<br/><i>Voice Activity Detection</i>"]
        G["Deepgram Nova<br/><b>Speech to Text (STT)</b>"]
        H["Amazon Bedrock<br/><b>Claude Haiku 4.5</b>"]
        I["Tool Use<br/><i>get_current_weather()</i>"]
        J["Deepgram Aura<br/><b>Text to Speech (TTS)</b>"]
    end

    B <-->|"Real-time Communication (WebRTC)"| Transport

    C -->|Voice input| E
    E --> F
    F --> G
    G -->|NLU query| H
    H -->|Function call| I
    I -->|Result| H
    H -->|NLG response| J
    J -->|Voice output| D

    style Users fill:#fafafa,stroke:#ddd,color:#888
    style Transport fill:#fafafa,stroke:#ddd,color:#888
    style Pipeline fill:#fafafa,stroke:#ddd,color:#888
    style A fill:#333,color:#fff
    style B fill:#fff,stroke:#ddd
    style C fill:#fff,stroke:#ddd
    style D fill:#fff,stroke:#ddd
    style E fill:#fff,stroke:#ddd
    style F fill:#fff,stroke:#ddd
    style G fill:#fff,stroke:#ddd
    style H fill:#fff,stroke:#ddd
    style I fill:#fff,stroke:#ddd
    style J fill:#fff,stroke:#ddd
Loading

Customize the bot

The server/bot.py file contains your Pipecat bot. To customize it, look for the comment #### Customize bot prompt here! Update "content". That messages variable is used by the bot's context manager, which stores the conversation between the bot and the user. Change the content property of that first message to update your bot's system prompt.

Next, you'll almost certainly want to use function calling to extend your bot's functionality. Search for the comments #### Customize function here! to see how this bot can answer questions about the weather (using fake data). Read more about function calling in the Pipecat docs page about it.

Add AWS functionality

Coming soon!

Deploy to Pipecat Cloud

For the hackathon, you can perform your live demo by running the bot locally on your computer. To deliver a hosted version, use Pipecat Cloud.

Tip

You can deploy via this Github Action

  • Login to DockerHub:
docker login
# create a personal access token at settings/personal-access-tokens/create
  • Setup Pipecat CLI:
uv tool install "pipecat-ai-cli[tail]"
pipecat cloud auth login
pipecat cloud secrets image-pull-secret my-image-pull-secret https://index.docker.io/v1/ # use your dockerhub username, but for password, put your personal access token
pipecat cloud docker build-push -u <your_dockerhub_username>
nvim pcc-deploy.toml # and add your username to the image path
pipecat cloud secrets set --file .env aws-deepgram-2026-03-secrets
pipecat cloud deploy # pcc_deploy.toml is configured to use my-image--pull-secret
export PCC_API_KEY=pk_...
export YOUR_AGENT=...
curl -s --request POST \
--url "https://api.pipecat.daily.co/v1/public/${YOUR_AGENT}/start" \
--header "authorization: Bearer $PCC_API_KEY" \
--header 'content-type: application/json' \
--data '{
  "createDailyRoom": true,
  "dailyRoomProperties": { "start_video_off": true }
}' |jq -r '"\n" + (.dailyRoom | tostring) + "?t=" + (.dailyToken | tostring)
  • Configure and deploy frontend:

Coming soon!

Troubleshooting

required .env variables for Server

required .env variables for Client