Thanks — I can help with that. Are you asking how to use the LiveKit Agent Builder in the Cloud dashboard, or do you mean exporting the agent to code (the downloadable Python project)? Which of these would you like help with, or do you want a quick walkthrough of the whole builder? If you want a quick walkthrough of the builder, here’s a concise step-by-step you can follow: 1. Access the Agent Builder ◦ Open your LiveKit Cloud project and go to the Agents dashboard, then click "Deploy new agent". 1. Configure agent basics ◦ Set an Agent Name (used for explicit dispatch). ◦ Fill the Instructions (system prompt) to define agent behavior. You can use variables like

{{metadata.user_name}}

. ◦ Optionally set a Welcome Greeting that plays when a user joins. 1. Select models ◦ Choose STT, LLM, and TTS models supported by LiveKit Inference (for example Deepgram, GPT-family models, ElevenLabs). 1. Add actions (tools) ◦ Add HTTP Tools to call external APIs (GET/POST/PUT/PATCH/DELETE). You can include headers, params, and mark tools as silent. ◦ Add Client Tools to call frontend RPC methods. ◦ Optionally connect an MCP server so the agent can discover additional tools. 1. Use variables and secrets ◦ Access metadata with

{{metadata.field_name}}

. ◦ Use secrets stored in the builder with

{{secrets.SECRET_NAME}}

for API keys in tool headers. 1. Preview ◦ Use the built-in live preview to talk to your agent in the browser. Preview sessions use Inference credits but don’t count as production sessions. 1. Deploy ◦ Click “Deploy agent” to make it a production LiveKit Cloud agent (compatible with telephony, observability, and dispatch). 1. Test and monitor ◦ Use the agent’s “Test Agent” public sandbox link for demos. ◦ Monitor production sessions in Agent Insights / Sessions dashboard. 1. Export to code (optional) ◦ Click “Download code” to get a Python project ZIP (includes AGENTS.md and MCP server integration). If you deploy the code version, delete the builder agent to avoid duplicate requests. Limitations to keep in mind: • Builder does not support Workflows, handoffs/tasks, virtual avatars/vision, realtime models/plugins, or unit testing. Billing note: • There’s no extra cost to use the builder; normal LiveKit Cloud quotas and limits apply. Tell me which step you want to dive into (examples: configuring HTTP tools, writing the agent instructions, setting up secrets, or exporting and running the Python project) and I’ll guide you through it with exact actions. Sources: Agent Builder | LiveKit Documentation | Announcing Agent Builder: a pathway to code | LiveKit DevDay | LiveKit