Overview
LiveKit Agents supports modular components for structuring your agent's logic and behavior. Use sessions, workflows, and tools to build custom logic for handling conversations, automations, and external integrations.
Logic and structure components
Use core components to structure your agent logic, including sessions, workflows, customization points, and external integrations. Build simple single-agent applications, or combine these components for complex, multi-agent workflows.
| Component | Description | Use cases |
|---|---|---|
| Agent sessions | Orchestrate input collection, pipeline management, and output delivery. The main orchestrator for your voice AI app. | Single-agent apps, session lifecycle management, and room I/O configuration. |
| Tasks & task groups | Create focused, reusable units that perform specific objectives and return typed results. Tasks run inside agents and take temporary control until completion. | Consent collection, structured data capture, and multi-step processes with task groups. |
| Workflows | Model repeatable patterns with agents, handoffs, and tasks for complex voice AI systems. | Multi-persona systems, conversation phase management, and specialized agent routing. |
| Tool definition & use | Extend agent capabilities with custom functions callable by the LLM for external actions and data access. | API integrations, frontend RPC calls, and triggering agent handoffs. |
| Pipeline nodes & hooks | Customize agent behavior at pipeline processing points with custom STT, LLM, TTS, and lifecycle hooks. Override nodes to modify input, output, or add custom logic. | Custom providers, output modification, and pronunciation control. |
| Turn detection & interruptions | Manage conversation flow with turn detection, interruption handling, and manual turn control. | Natural conversation timing, interruption management, and push-to-talk interfaces. |
| Agents & handoffs | Define distinct reasoning behaviors and transfer control between agents when different capabilities are needed. | Role-based agents, model specialization, and permission management. |
| External data & RAG | Connect agents to external data sources, databases, and APIs for RAG and data operations. Load initial context, perform RAG lookups, and integrate with external services. | Knowledge base search, user profile loading, and database operations. |
In this section
Read more about each component.
Agent sessions
Main orchestrator for input collection, pipeline management, and output delivery.
Tasks & task groups
Focused units that perform specific objectives and return typed results.
Workflows
Model repeatable patterns with agents, handoffs, and tasks.
Tool definition & use
Custom functions callable by the LLM for external actions.
Pipeline nodes & hooks
Customize behavior at pipeline processing points.
Turn detection & interruptions
Manage conversation flow with turn detection and interruption handling.
Agents & handoffs
Define distinct agents and transfer control between them.
External data & RAG
Connect to external data sources, databases, and APIs.