LiveKit just killed the silent voice agent. Async tools hand control back before they finish.

Voice AI agents have a brutal UX problem.

When a tool runs — booking a flight, querying a database, calling an API — the user gets dead silence. No confirmation. No progress. Just a frozen mic.

LiveKit Agents 1.6.0 ships the fix: first-class async tools via `ctx.update()`.

Here's the new contract:

A long-running tool fires an immediate update — "Searching flights to Tokyo, this'll take a minute" — then hands control back to the LLM.

The agent keeps talking. The tool keeps working in the background.

When it finishes, the result streams in as a deferred reply when the agent is idle.

They also shipped `ctx.with_filler()` — configurable filler phrases on a timer. "Almost there." "Hang tight." Rotate them, set the delay, set the interval. No more awkward pauses.

The code pattern:

await ctx.update("Searching flights. This'll take a minute.")
# tool runs in background, agent keeps talking
await asyncio.sleep(40)
return f"Booked! Confirmation FL-{id}."

This is tool-use done right. The LLM doesn't block on I/O. The user doesn't sit in silence. The tool manages its own lifecycle.

If you're building voice agents and your tools still block the conversation, you're shipping a product that feels broken. Rotate to this pattern.

11K stars. Apache 2.0. LiveKit Agents 1.6.0.