LiveKit Agents Docs • LiveKit Cloud • Blog
This example demonstrates an full workflow of an AI agent that makes outbound calls. It uses LiveKit SIP and Python Agents Framework.
It can use a pipeline of STT, LLM, and TTS models, or a realtime speech-to-speech model. (such as ones from OpenAI and Gemini).
This example builds on concepts from the Outbound Calls section of the docs. Ensure that a SIP outbound trunk is configured before proceeding.
This example demonstrates the following features:
- Making outbound calls
- Detecting voicemail
- Looking up availability via function calling
- Transferring to a human operator
- Detecting intent to end the call
- Uses Krisp background voice cancellation to handle noisy environments
Clone the repository and install dependencies to a virtual environment:
git clone https://github.com/livekit-examples/outbound-caller-python.git
cd outbound-caller-python
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
python agent.py download-filesSet up the environment by copying .env.example to .env.local and filling in the required values:
LIVEKIT_URLLIVEKIT_API_KEYLIVEKIT_API_SECRETOPENAI_API_KEYSIP_OUTBOUND_TRUNK_IDDEEPGRAM_API_KEY- optional, only needed when using pipelined modelsCARTESIA_API_KEY- optional, only needed when using pipelined models
Run the agent:
python3 agent.py devNow, your worker is running, and waiting for dispatches in order to make outbound calls.
You can dispatch an agent to make a call by using the lk CLI:
lk dispatch create \
--new-room \
--agent-name outbound-caller \
--metadata '{"phone_number": "+1234567890", "transfer_to": "+9876543210}'