Today we’re introducing Agent Builder, a browser-based experience that lets you create, test, and deploy production-grade LiveKit voice agents in minutes, with zero local setup.

These are not prototypes or sandbox toys. Every agent you build in Agent Builder is a real, fully functional LiveKit Agent running on LiveKit Cloud, connected to the exact same voice pipeline, phone numbers, observability, and client SDKs you use in production. From the very first call, you’re shipping the real thing.

How it works

Using Agent Builder is straightforward. Start with a single, clear system prompt for what you want your agent to do. LiveKit provides a production‑ready voice pipeline out of the box with built-in background noise cancellation, preemptive generation, and multilingual end‑of‑turn detection.

You can have your agent greet users with custom instructions, capture and export transcripts, and pass job metadata to shape behavior and logging. Every change autosaves with version history, so it’s easy to iterate quickly in your browser.

Connect real tools in seconds

Extend your agent with HTTP tools to pull or push data. Define GET or POST endpoints that your agent can call during conversations. For example:

• Fetching customer records from a CRM
• Retrieving order status from an API
• Logging form responses to a webhook

Configure each tool with a URL, optional headers, and request body, and the agent decides when to invoke it based on the context of the conversation. Responses are passed back into the agent's context so it can use real-time data to inform its replies.

Test live, deploy instantly

Click the preview button and talk to your agent right in the browser. When you’re ready, deploy with one click to LiveKit Cloud. There’s nothing “demo‑only” here: Agent Builder creates real agents defined in code using the LiveKit Agents framework. You can inspect the live code at any time and download it as a zip file to continue in your repo, add tests, and customize as you see fit. It’s a fast on‑ramp, not a detour.

Built to ship fast and iterate faster

Agent Builder is intentionally simple. It’s ideal for:

• Teammates evaluating voice with a quick proof of concept
• Engineers who want a guided first mile before moving to the SDKs
• Builders migrating from other platforms who prefer a prompt‑first model.

Because it uses the same core primitives as the rest of LiveKit, it works with phone numbers, client SDKs, and observability from day one.

Under the hood, LiveKit Inference gives you instant access to a curated set of speech and language models, and the in-browser preview lets you test different model combinations—from STT providers like Deepgram and AssemblyAI to LLMs like GPT-4.1 and Gemini 2.5 Flash, to TTS voices from ElevenLabs and Cartesia—without writing a line of code or redeploying. You can hear the difference between models in real time, compare latency and voice quality, and pick what works best for your use case before you commit to production.

Designed as a pathway to code

Agent Builder is not a low‑code island. It is designed as a pathway to code, and we are investing in tooling that makes the transition from prototype to production‑grade SDK implementation seamless.

We recently shipped a Docs MCP server so tools like Cursor and Claude can answer “how do I…?” with precise, contextual guidance drawn straight from LiveKit documentation. Point your assistant at your project and our docs, and you will get inline suggestions that match the APIs you are actually using.

We also introduced repository‑level AGENTS.md files that describe an agent’s purpose, capabilities, tools, and environment in a format coding assistants understand. This gives teammates and AI assistants a shared source of truth, so tasks like adding a tool, tweaking prompts, or exporting code stay consistent as your project grows.

Together, Agent Builder, Docs MCP, and AGENTS.md help any developer—regardless of starting skill level—turn an initial voice prototype into a reliable, production‑ready agent built with the full LiveKit Agents framework.

What’s next

Agent Builder is not meant to replace building agents in code, but it is a core part of how we improve the end‑to‑end developer experience of building on LiveKit. Upcoming features include:

• Support for client tools via RPC
• Realtime APIs for speech‑to‑speech models
• Cost estimates to help you plan for scale
• A streamlined export‑to‑GitHub workflow

Our goal is to make it easy to start in the browser, then grow into whatever level of sophistication your product requires.

Get started today

Agent Builder is available to all LiveKit Cloud projects. Create an agent with a simple prompt, try it live in your browser, deploy with one click, then export the code when you’re ready to keep building in your preferred SDK and repository.

If you’ve been curious about voice agents but haven’t made it past the playground, Agent Builder is the fastest way to turn curiosity into something real and set yourself up to scale tomorrow. Try it today and let us know how it goes.