livekit-skills

Initialize a project with the LiveKit CLI using a starter template like agent-starter-python or agent-starter-node, set your LIVEKIT_URL, LIVEKIT_API_KEY, and LIVEKIT_API_SECRET credentials, then run the agent in dev mode with lk agent dev.

Design handoffs around natural conversation boundaries where one agent transfers control to another, such as moving from intake to a billing specialist. Consult the LiveKit documentation via lk docs for current implementation details on handoffs and tasks.

Yes, you can run livekit-server locally in dev mode with default credentials devkey and secret, or deploy it in production via Docker, Kubernetes, or VMs. Self-hosting removes Cloud tier limits but requires configuring your own AI model provider API keys.

The Agents SDK has plugins for major providers including Deepgram for STT, OpenAI for LLM, and ElevenLabs or Cartesia for TTS. LiveKit Cloud also offers LiveKit Inference, which provides model access through your LiveKit credentials without separate API keys.

The LiveKit Agents SDK evolves faster than model training cycles, so memorized API signatures and defaults are frequently outdated. Always verify against live documentation using lk docs search, get-page, and code-search commands before writing code.

Yes, every agent implementation should include tests covering basic conversation flow, tool invocation, and error handling. Prompt and tool description changes silently alter behavior, so tests catch regressions before users encounter them in production.