> Fetch clean Markdown by appending `.md` to any page URL under https://signalwire.com/docs or requesting it with the HTTP header `Accept: text/markdown`. The root index at https://signalwire.com/docs/llms.txt lists the available documentation indexes.

# Agent

> LiveKit-compatible agent class that holds instructions, tools, and user data.

[agentsession]: /docs/server-sdks/reference/typescript/agents/livewire/agent-session

[agentbase]: /docs/server-sdks/reference/typescript/agents/agent-base

[functiontool]: /docs/server-sdks/reference/typescript/agents/livewire/function-tool

The `Agent<UserData>` class mirrors a LiveKit voice agent. It holds the system
prompt (instructions), a set of tools, and optional user data. When bound to an
[`AgentSession`][agentsession], the session translates this into a SignalWire
[`AgentBase`][agentbase] under the hood.

```typescript {11-14}
import { Agent, tool } from '@signalwire/sdk/livewire';

const getWeather = tool({
  description: 'Get the current weather for a city.',
  parameters: { city: { type: 'string' } },
  execute: (params) => {
    return `It is sunny in ${params.city}.`;
  },
});

const agent = new Agent({
  instructions: 'You are a helpful weather assistant.',
  tools: [getWeather],
});
```

## **Constructor**

```typescript {1-13}
new Agent<UserData>(options?: {
  instructions?: string;
  tools?: FunctionTool[];
  userData?: UserData;
  chatCtx?: unknown;
  stt?: unknown;
  tts?: unknown;
  llm?: unknown;
  vad?: unknown;
  turnDetection?: unknown;
  mcpServers?: unknown;
  allowInterruptions?: boolean;
  minEndpointingDelay?: number;
  maxEndpointingDelay?: number;
})
```

> **Info**
>
> The constructor options are source-compatible with LiveKit's `Agent`
> constructor, including pipeline fields like `stt`, `tts`, `vad`,
> `turnDetection`, and `mcpServers`. SignalWire handles speech, TTS, and VAD on
> the control plane, so those fields are accepted but logged as a no-op the
> first time they are supplied.

**`instructions`** `string` — default: ""

System prompt text. Mapped to the SignalWire prompt when the session starts.

---

**`tools`** `FunctionTool[]` — default: \[]

Array of [`FunctionTool`][functiontool] objects. Stored internally as a
name-keyed record; each tool is registered as a SWAIG function on the
underlying agent.

---

**`userData`** `UserData` — default: undefined

Arbitrary data attached to the agent. Accessible from tool handlers via
`RunContext.userData`.

---

**`llm`** `unknown`

Accepted for LiveKit parity. Pipeline mapping hint — SignalWire selects the
model from the SWML `ai.params`.

---

**`allowInterruptions`** `boolean`

Accepted for LiveKit parity. Forwarded to the SignalWire AI params when the
session starts.

---

**`minEndpointingDelay`** `number`

Accepted for LiveKit parity. Forwarded to the SignalWire AI params when the
session starts.

---

**`maxEndpointingDelay`** `number`

Accepted for LiveKit parity. Forwarded to the SignalWire AI params when the
session starts.

---

**`stt / tts / vad / turnDetection / mcpServers`** `unknown`

Accepted for LiveKit source compatibility. SignalWire handles these
capabilities on the control plane; passing a value logs a one-time
"no-op" advisory and does not affect behavior.

---

## **Properties**

**`instructions`** `string`

The system prompt text for the agent. Can be read or modified directly.

---

**`tools`** `Record<string, FunctionTool>`

Internal name-keyed map of registered tools.

---

**`userData`** `UserData | undefined`

The user data attached to this agent.

---

**`session`** `AgentSession | undefined`

The [`AgentSession`][agentsession] bound to this agent, if any. Set when
`session.start({ agent })` is called.

---

## **Example**

```typescript {13-17}
import { Agent, AgentSession, tool, defineAgent, runApp, JobContext } from '@signalwire/sdk/livewire';

const lookupAccount = tool({
  description: 'Look up a customer account by email.',
  parameters: { email: { type: 'string' } },
  execute: (params) => {
    return `Account found for ${params.email}: Premium tier.`;
  },
});

const agentDef = defineAgent({
  entry: async (ctx: JobContext) => {
    const agent = new Agent({
      instructions: 'You are a customer support agent. Help users with their accounts.',
      tools: [lookupAccount],
      userData: { department: 'support' },
    });

    const session = new AgentSession();
    await session.start({ agent });
    session.generateReply({ instructions: 'Greet the customer and ask how you can help.' });
  },
});

runApp(agentDef);
```

> LiveKit-compatible agent class that holds instructions, tools, and user data.