LiveWire | SignalWire

LiveWire is a compatibility layer that lets developers familiar with livekit-agents use the same class and function names while running on SignalWire infrastructure. Change your import path and your existing code runs on SignalWire with no other modifications.

1 // Before (livekit-agents)
2 import { Agent, AgentSession } from '@livekit/agents';
3 
4 // After (SignalWire LiveWire)
5 import { Agent, AgentSession } from '@signalwire/sdk/livewire';

SignalWire’s control plane handles STT, TTS, VAD, and turn detection automatically. Pipeline plugin parameters (stt, tts, vad, turnDetection) are accepted for API compatibility but are no-ops. LiveWire logs an informational message the first time each no-op parameter is used.

Quick Start

1 import {
2   Agent, AgentSession, JobContext,
3   tool, defineAgent, runApp,
4 } from '@signalwire/sdk/livewire';
5 
6 const lookupOrder = tool({
7   description: 'Look up the status of a customer order.',
8   parameters: { orderId: { type: 'string' } },
9   execute: (params) => {
10     return `Order ${params.orderId} shipped yesterday.`;
11   },
12 });
13 
14 const agentDef = defineAgent({
15   entry: async (ctx: JobContext) => {
16     const agent = new Agent({
17       instructions: 'You are a helpful customer support agent.',
18       tools: { lookupOrder },
19     });
20     const session = new AgentSession();
21     await session.start({ agent });
22     session.generateReply({ instructions: 'Greet the user.' });
23   },
24 });
25 
26 runApp(agentDef);

Namespace Aliases

LiveWire provides namespace objects that mirror common livekit-agents import paths:

Alias	Contents	Mirrors
`voice`	`Agent`, `AgentSession`	`livekit.agents.voice`
`llm`	`tool`, `handoff`, `ToolError`, `ChatContext`	`livekit.agents.llm`
`cli`	`runApp`	`livekit.agents.cli`
`inference`	`STT`, `LLM`, `TTS`	`livekit.agents.inference`

1 import { voice, inference, cli } from '@signalwire/sdk/livewire';
2 
3 const agent = new voice.Agent({ instructions: 'Hello' });
4 const session = new voice.AgentSession({
5   llm: new inference.LLM('gpt-4o'),
6 });