> ## Documentation Index
> Fetch the complete documentation index at: https://ai.aidalinfo.fr/llms.txt
> Use this file to discover all available pages before exploring further.

# Agent class

> Configuration, methods, and execution options for an AI Kit agent.

```ts theme={null}
import { Agent, type AgentConfig } from "@ai_kit/core";
```

## `AgentConfig`

| Property       | Type            | Default                         | Description                                                |
| -------------- | --------------- | ------------------------------- | ---------------------------------------------------------- |
| `name`         | `string`        | required                        | Agent identifier (logging, telemetry).                     |
| `instructions` | `string`        | `undefined`                     | Default system prompt (override with `system`).            |
| `model`        | `LanguageModel` | required                        | Model from the AI SDK.                                     |
| `tools`        | `AgentTools`    | `undefined`                     | Tools exposed by default (`ToolSet` or provider-specific). |
| `telemetry`    | `boolean`       | `false`                         | Enables OTEL / Langfuse export by default.                 |
| `loopTools`    | `boolean`       | `false`                         | Enables the automatic tool loop.                           |
| `maxStepTools` | `number`        | `20` (`DEFAULT_MAX_STEP_TOOLS`) | Max tool executions before the loop stops.                 |

## Constructor

```ts theme={null}
const agent = new Agent(config: AgentConfig);
```

### Instance methods

#### `agent.generate(options)`

```ts theme={null}
generate<OUTPUT = never, PARTIAL = never, STATE extends RuntimeState = RuntimeState>(
  options: AgentGenerateOptions<OUTPUT, PARTIAL, STATE>
): Promise<AgentGenerateResult<OUTPUT>>;
```

Options extend AI SDK `generateText` parameters (`prompt` or `messages`) with:

* `system?: string` – overrides the default instructions.
* `structuredOutput?: Output.Output<OUTPUT, PARTIAL>` – schema-based generation (Zod).
* `runtime?: RuntimeStore<STATE>` – shared runtime (see [`RuntimeStore`](/en/api-reference/runtime-store)).
* `telemetry?: AgentTelemetryOverrides` – merges `functionId`, `metadata`, `recordInputs`, `recordOutputs`.
* `loopTools?: boolean` / `maxStepTools?: number` – per-call overrides for the tool loop.

Returns a `GenerateTextResult` enriched with `loopTool?: boolean`.

#### `agent.stream(options)`

```ts theme={null}
stream<OUTPUT = never, PARTIAL = never, STATE extends RuntimeState = RuntimeState>(
  options: AgentStreamOptions<OUTPUT, PARTIAL, STATE>
): Promise<AgentStreamResult<PARTIAL>>;
```

Additional properties:

* `textStream`, `fullStream`, `text`, `response`, `usage`, `steps`.
* `experimental_partialOutputStream` when `structuredOutput` is provided.
* `toAIStreamResponse()` / `toDataStreamResponse()` to integrate with HTTP streaming responses.

Structured output automatically switches to the structured pipeline when supported by the model.

#### `agent.withTelemetry(enabled?: boolean)`

Enable or disable telemetry by mutating the instance. Returns `this` for chaining.

```ts theme={null}
agent.withTelemetry();      // enable
agent.withTelemetry(false); // disable
```

## Utility types

* `AgentTools` – union between AI SDK `ToolSet` and provider dictionaries.
* `AgentGenerateResult<T>` – alias of `GenerateTextResult` + `loopTool`.
* `AgentStreamResult<T>` – alias of `StreamTextResult` + `loopTool`.
* `AgentTelemetryOverrides` – `functionId`, `metadata`, `recordInputs?`, `recordOutputs?`.

## Runtime handling

When `runtime` is provided:

* a `RuntimeStore` snapshot is created for the call duration;
* `onCleanup` handlers run at the end (success, error, or cancellation);
* `RuntimeStore.mergeExperimentalContext` merges manually passed `experimental_context` with the runtime state.

Without a runtime, the agent simply wraps the AI SDK behaviour.