---
title: "Text generation"
description: "Send prompts to Workers AI models with Laravel AI agents. Set the model, token cap and timeout."
---

> Documentation Index
> Fetch the complete documentation index at: https://laravel-cf-workersai.meirdick.com/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Text generation

## Prompt an agent

For a one-off call, use the `agent()` helper:

```php
use function Laravel\Ai\agent;

$response = agent('You are a helpful assistant.')
    ->prompt('Name three uses for a paperclip.', provider: 'workers-ai');

echo $response->text;
```

For an agent class, implement `Laravel\Ai\Contracts\Agent` and use the `Promptable` trait. Add `#[Provider('workers-ai')]` so the agent always runs on Workers AI.

```php
use Laravel\Ai\Attributes\Provider;
use Laravel\Ai\Contracts\Agent;
use Laravel\Ai\Promptable;

#[Provider('workers-ai')]
class SupportAgent implements Agent
{
    use Promptable;

    public function instructions(): string
    {
        return 'You answer support questions in two sentences.';
    }
}

$response = (new SupportAgent)->prompt('How do I reset my password?');
```

## Choose a model

The package uses these models when you name none:

| Role | Default model | Config key |
| --- | --- | --- |
| Default | `@cf/meta/llama-3.3-70b-instruct-fp8-fast` | `models.text.default` |
| `#[UseCheapestModel]` | `@cf/meta/llama-3.2-3b-instruct` | `models.text.cheapest` |
| `#[UseSmartestModel]` | `@cf/moonshotai/kimi-k2.6` | `models.text.smartest` |

To pick a model for one call, pass `model:`. For a whole agent, add `#[Model('@cf/...')]`. To change the defaults, use the config keys above. See [Models](/docs/guides/models/).

## Set the token cap

When a request has no token cap, Cloudflare stops the completion at 256 tokens, which cuts most structured output off in the middle of the JSON. So the package sends `max_completion_tokens: 4096` by default.

Change the default for all agents with `default_max_tokens`:

```php
'workers-ai' => [
    // ...
    'default_max_tokens' => 2048,
],
```

You can set it to `null` to send no cap, but avoid that. On 2026-09-09 on one account, `@cf/zai-org/glm-5.3-flash` with no cap ran to 8,190 tokens in 369 seconds.

Set a cap for one agent with `#[MaxTokens]`:

```php
use Laravel\Ai\Attributes\MaxTokens;

#[MaxTokens(500)]
class ShortAnswerAgent implements Agent { /* ... */ }
```

A per-call or per-agent value wins over `default_max_tokens`, which wins over the Cloudflare default.

## Set the timeout

Laravel AI uses a 60-second timeout, and large models, structured output and reasoning models can need more. To raise it:

```php
use Laravel\Ai\Attributes\Timeout;

#[Timeout(120)]
class ExtractionAgent implements Agent { /* ... */ }

// or per call, with any agent:
$agent = new ExtractionAgent;
$agent->prompt('...', provider: 'workers-ai', timeout: 120);
```

The package doesn't retry a request that times out. It fails at once, because a retry would run as long again. See [Errors and retries](/docs/guides/errors-and-retries/).

## Temperature and top-p

The package sends `temperature` and `top_p` when Laravel AI sets them, which you do with `#[Temperature]` and `#[TopP]` on the agent.

## Send extra request fields

To add fields to the request body, implement `Laravel\Ai\Contracts\HasProviderOptions` on the agent. The package merges the array you return into the body.

```php
use Laravel\Ai\Contracts\HasProviderOptions;
use Laravel\Ai\Enums\Lab;

class MyAgent implements Agent, HasProviderOptions
{
    use Promptable;

    public function instructions(): string
    {
        return 'You are a helpful assistant.';
    }

    public function providerOptions(Lab|string $provider): array
    {
        return $provider === 'workers-ai'
            ? ['seed' => 42]
            : [];
    }
}
```

For this package, `$provider` is a string, not a `Lab` case. The package doesn't check the fields, so read the model's page in the Cloudflare catalog to see which ones it accepts.

> **Note**
>
> Keys from `providerOptions()` replace keys the package builds, such as `reasoning_effort`. The one exception is `session_affinity`, which the package removes from the body; set it in config instead.

Source: https://laravel-cf-workersai.meirdick.com/docs/guides/text-generation/index.mdx
