LiveKit Agents

Operator-focused playbook for building and running LiveKit Agents (voice/video AI agents) with the LiveKit Agent Server.

Quick Navigation

• Installation & setup: references/installation.md
• Getting started: references/get-started.md
• Components (Agents UI): references/components.md
• Prompting & evaluation: references/prompting-and-testing.md
• Multimodality: references/multimodality.md
• Logic (sessions/tasks/workflows/tools): references/logic-and-structure.md
• Prebuilt tasks & tools: references/prebuilt.md
• Agent Server (jobs, lifecycle, options): references/agent-server.md
• Models (realtime/inference/integrations): references/models.md
• CLI + events/errors: references/reference.md

When to Use

Use this skill when you need to:

• Build a voice agent or multimodal agent on LiveKit
• Structure agent logic (sessions, tasks, workflows) and tool calling
• Run agents in production via the Agent Server (job lifecycle, dispatch)
• Choose/operate models (realtime, inference) and connect STT/TTS/LLMs
• Design prompts and run tests/evaluations

Core Recipes

Recipe: Voice agent quickstart (minimal)

• Goal: get an agent into a LiveKit room, capture audio, and respond.
• Use: references/get-started.md

Recipe: Production Agent Server (jobs + lifecycle)

• Goal: run agents reliably, handle startup mode, dispatch, and job lifecycle.
• Use: references/agent-server.md

Recipe: Tool calling (safe + debuggable)

• Goal: define tools, validate inputs/outputs, and avoid prompt/tool drift.
• Use: references/logic-and-structure.md

Critical Prohibitions

• Do not mirror vendor docs verbatim; keep references actionable.
• Do not hardcode secrets (API keys, LiveKit keys) into prompts or code snippets.
• Do not rely on silent fallbacks for model/provider selection; fail fast.

Links

• Documentation: https://docs.livekit.io/agents/
• Python SDK (reference): https://docs.livekit.io/reference/python/livekit/agents/index.html
• GitHub (agents): https://github.com/livekit/agents