Overview
LiveKit Agents includes a number of ready-made tasks components that you can use to reliably solve recurring problems in voice AI. These tasks have been tested and tuned based on real-world scenarios and can help you rapidly build your own voice AI applications.
Tasks are a way to break your agent's logic into well-scoped, reliable chunks. In addition to the prebuilt tasks, you can also easily build your own custom tasks to perform any job you need. For more information, see Tasks and task groups.
These tasks can be found in the livekit.agents.beta.workflows module.
LiveKit Agents for Python contains the following prebuilt tasks:
| Task | Description |
|---|---|
| GetEmailTask | Collect and validate an email address from the user. Handles noisy voice transcription and spoken patterns. |
| GetAddressTask | Collect and validate a complete mailing address. Supports international formats and spoken input. |
| GetDtmfTask | Collect keypad (DTMF) or spoken digits from callers. For IVR menus, PIN entry, and digit capture. |
| WarmTransferTask | Execute an agent-assisted warm transfer. Dials the supervisor via SIP, plays hold music, and hands off context. |
Usage
Await a prebuilt task from within your agent, typically from a function tool. The task runs until it completes and returns a result. You can run tasks inside a task group to chain or sequence them.
Customization
Prebuilt tasks support customization so you can adapt them to your use case. Use the extra_instructions parameter to append instructions to a task's default behavior. The task's LLM receives both the built-in instructions and your extra text, so you can steer prompts, add context, or change when to use optional tools. Some tasks also accept a tools parameter so you can add or substitute function tools. See each task's parameters section for details.