blaspsoft / forerunner
A Laravel package that provides an elegant, migration-inspired API for defining JSON schemas that ensure your LLM responses are perfectly structured every time.
Requires
- php: ^8.2
- illuminate/contracts: ^12.0|^13.0
- illuminate/support: ^12.0|^13.0
Requires (Dev)
- laravel/pint: ^1.13
- orchestra/testbench: ^10.0|^11.0
- pestphp/pest: ^3.0|^4.0
- pestphp/pest-plugin-laravel: ^3.0|^4.0
- phpstan/phpstan: ^2.0
Suggests
None
Provides
None
Conflicts
None
Replaces
None
README
Forerunner - Build structured LLM outputs the Laravel way
A Laravel package that provides an elegant, migration-inspired API for defining JSON schemas that ensure your LLM responses are perfectly structured every time.
Installation
You can install the package via composer:
composer require blaspsoft/forerunner:^0.4
Note: This is a pre-release version (0.x). The API may change as we gather feedback and iterate towards 1.0.0.
The package will automatically register its service provider.
Upgrading from 0.1.x
If you're upgrading from version 0.1.x, please note there are breaking changes. See the v0.2.0 release notes for a complete migration guide.
Quick Start
Using the Artisan Command
Generate a new structure class:
php artisan make:struct UserProfile
This creates a structure class at app/Structures/UserProfile.php:
<?php namespace App\Structures; use Blaspsoft\Forerunner\Schema\Struct; use Blaspsoft\Forerunner\Schema\Property; class UserProfile { public static function schema(): Struct { return Struct::define('user_profile', 'Description of user_profile', function (Property $property) { $property->string('example_field'); // Add your fields here $property->strict(); // All fields required + no additional properties }); } }
Basic Usage
Define a schema using the Struct class or Schema facade:
use Blaspsoft\Forerunner\Schema\Struct; use Blaspsoft\Forerunner\Schema\Property; $schema = Struct::define('User', 'A user schema', function (Property $property) { $property->string('name', 'The user\'s full name')->required(); $property->string('email', 'The user\'s email address')->required(); $property->int('age', 'The user\'s age')->min(0)->max(150); $property->boolean('is_active', 'Is the user account active?')->default(true); })->toArray();
Or using the facade for a cleaner syntax:
use Blaspsoft\Forerunner\Facades\Schema; use Blaspsoft\Forerunner\Schema\Property; $schema = Schema::define('User', 'A user schema', function (Property $property) { $property->string('name', 'The user\'s full name')->required(); $property->string('email', 'The user\'s email address')->required(); })->toArray();
Available Field Types
String Fields
$property->string('username', 'The username') ->minLength(3) ->maxLength(50) ->pattern('^[a-zA-Z0-9_]+$') ->required();
Integer Fields
$property->int('age', 'User age') ->min(0) ->max(150) ->default(18); // Alias $property->integer('count');
Float/Number Fields
$property->float('price', 'Product price') ->min(0.0) ->max(9999.99); // Alias $property->number('rating')->min(0)->max(5);
Boolean Fields
$property->boolean('is_active', 'Account status') ->default(true); // Alias $property->bool('verified');
Array Fields
// Simple array $property->array('tags', 'User tags') ->items('string') ->minItems(1) ->maxItems(10); // Array of objects $property->array('addresses')->items('object', function (Property $item) { $item->string('street')->required(); $item->string('city')->required(); $item->string('zip')->required(); });
Enum Fields
$property->enum('role', ['admin', 'user', 'guest'], 'User role') ->default('user'); $property->enum('status', ['draft', 'published', 'archived']);
Object Fields
$property->object('address', function (Property $nested) { $nested->string('street', 'Street address')->required(); $nested->string('city', 'City name')->required(); $nested->string('zip', 'ZIP code')->required(); $nested->object('coordinates', function (Property $coords) { $coords->float('latitude')->required(); $coords->float('longitude')->required(); }); }, 'User address');
Field Constraints
String Constraints
$property->string('username') ->minLength(3) // Minimum length ->maxLength(50) // Maximum length ->pattern('^[a-zA-Z0-9]+$') // Regex pattern ->required(); // Mark as required
Numeric Constraints
$property->int('age') ->min(0) // Minimum value ->max(150) // Maximum value ->default(18); // Default value
Array Constraints
$property->array('tags') ->items('string') // Type of array items ->minItems(1) // Minimum array length ->maxItems(10); // Maximum array length
General Constraints
$property->string('field') ->required() // Mark as required ->optional() // Mark as optional (default) ->default('value') // Set default value ->description('Field description'); // Add description
Advanced Features
Helper Methods for Common Formats
Forerunner provides convenient helper methods for commonly used field formats:
// Email field with automatic format validation $property->email('email')->required(); // URL field $property->url('website'); // UUID field $property->uuid('id')->required(); // Date-time field (ISO 8601) $property->datetime('created_at'); // Date field $property->date('birth_date'); // Time field $property->time('start_time'); // IPv4 address $property->ipv4('ip_address'); // IPv6 address $property->ipv6('ipv6_address'); // Hostname $property->hostname('server_name');
String Format Validation
You can also set custom formats on string fields:
$property->string('email')->format('email'); $property->string('website')->format('uri'); $property->string('id')->format('uuid');
Supported formats: email, uri, url, uuid, date, date-time, time, ipv4, ipv6, hostname, and more.
Nullable Fields
Mark fields as nullable to allow both the specified type and null:
$property->string('middle_name')->nullable(); // Generates: {"type": ["string", "null"]} $property->object('address', function (Property $nested) { $nested->string('street')->required(); $nested->string('city')->required(); })->nullable(); // Generates: {"type": ["object", "null"], "properties": {...}}
Unique Array Items
Ensure array items are unique:
$property->array('tags') ->items('string') ->uniqueItems();
Additional Properties Control
Control whether objects can have properties not defined in the schema:
// Allow additional properties $property->additionalProperties(true); // Disallow additional properties $property->additionalProperties(false); // This is the default // Or use the convenient strict() helper $property->strict(); // Disallows additional properties AND marks all fields as required
Strict Mode for LLM APIs
The strict() method marks every field as required and disallows additional properties, at every level of nesting (nested objects and array item objects included). It doesn't matter where in the callback you call it.
$struct = Struct::define('User', 'A user schema', function (Property $property) { $property->strict(); $property->string('fullname'); $property->email('email'); $property->object('address', function (Property $address) { $address->string('city'); // also required }); });
When a struct is strict, toArray() returns OpenAI's named wrapper (name, description, strict, schema) for backwards compatibility. Use toSchema() for the plain JSON schema, or one of the provider methods for a ready-to-send payload. You don't need strict() for those: each provider method applies that provider's rules itself.
Schema Metadata
Add metadata to your schemas:
$property->title('User Schema'); $property->description('Schema for user data validation'); $property->schemaVersion('https://json-schema.org/draft/2020-12/schema'); $property->id('https://example.com/schemas/user.json');
Schema Identifier ($id)
The id() method sets a unique identifier (URI) for your schema, following the JSON Schema specification. This is useful for referencing schemas and establishing canonical identifiers:
$schema = Struct::define('User', 'A user schema', function (Property $property) { $property->id('https://example.com/schemas/user.json'); $property->string('name')->required(); $property->int('age'); })->toArray();
This generates:
{
"$id": "https://example.com/schemas/user.json",
"type": "object",
"properties": {
"name": { "type": "string" },
"age": { "type": "integer" }
},
"required": ["name"],
"additionalProperties": false
}
The $id is optional and only included in the output when explicitly set.
You can also add titles to individual fields:
$property->string('email') ->title('Email Address') ->description('User\'s primary email address') ->format('email') ->required();
Complete Advanced Example
use Blaspsoft\Forerunner\Schema\Struct; use Blaspsoft\Forerunner\Schema\Property; $schema = Struct::define('AdvancedUser', 'Comprehensive user data structure', function (Property $property) { // Schema metadata $property->schemaVersion(); $property->id('https://example.com/schemas/advanced-user.json'); $property->title('Advanced User Schema'); // Helper methods $property->uuid('id')->required(); $property->email('email')->required(); $property->url('website')->nullable(); $property->datetime('created_at')->required(); // Nullable nested object $property->object('profile', function (Property $profile) { $profile->string('bio')->maxLength(500); $profile->string('avatar_url')->format('uri'); })->nullable(); // Array with unique items $property->array('tags') ->items('string') ->uniqueItems() ->minItems(1) ->maxItems(10); // Advanced field configuration $property->string('username') ->title('Username') ->description('Unique username for the account') ->minLength(3) ->maxLength(30) ->pattern('^[a-zA-Z0-9_]+$') ->required(); // Call strict() after all fields to mark them all as required $property->strict(); // Disallow extras + mark all defined fields as required })->toArray();
This generates:
{
"$schema": "https://json-schema.org/draft/2020-12/schema",
"$id": "https://example.com/schemas/advanced-user.json",
"type": "object",
"title": "Advanced User Schema",
"description": "Comprehensive user data structure",
"properties": {
"id": {
"type": "string",
"format": "uuid"
},
"email": {
"type": "string",
"format": "email"
},
"website": {
"type": ["string", "null"],
"format": "uri"
},
"created_at": {
"type": "string",
"format": "date-time"
},
"profile": {
"type": ["object", "null"],
"properties": {
"bio": {
"type": "string",
"maxLength": 500
},
"avatar_url": {
"type": "string",
"format": "uri"
}
}
},
"tags": {
"type": "array",
"items": {
"type": "string"
},
"uniqueItems": true,
"minItems": 1,
"maxItems": 10
},
"username": {
"type": "string",
"title": "Username",
"description": "Unique username for the account",
"minLength": 3,
"maxLength": 30,
"pattern": "^[a-zA-Z0-9_]+$"
}
},
"required": ["id", "email", "created_at", "username"],
"additionalProperties": false
}
Complex Examples
User Profile with Nested Objects
use Blaspsoft\Forerunner\Schema\Struct; use Blaspsoft\Forerunner\Schema\Property; $schema = Struct::define('UserProfile', 'A complete user profile schema', function (Property $property) { $property->string('name', 'The user\'s full name') ->minLength(1) ->maxLength(100) ->required(); $property->string('email', 'The user\'s email') ->pattern('^[^\s@]+@[^\s@]+\.[^\s@]+$') ->required(); $property->int('age', 'The user\'s age') ->min(0) ->max(150); $property->boolean('is_active', 'Is the account active?') ->default(true); $property->array('tags', 'User tags') ->items('string') ->minItems(0) ->maxItems(10); $property->object('address', function (Property $address) { $address->string('street', 'Street name')->required(); $address->string('city', 'City name')->required(); $address->string('state', 'State/Province')->required(); $address->string('zip', 'ZIP/Postal code')->required(); $address->string('country', 'Country code')->required(); }, 'User\'s address'); $property->enum('role', ['admin', 'moderator', 'user'], 'User role') ->default('user'); })->toArray();
Blog Post with Comments
$schema = Struct::define('BlogPost', 'A blog post with author and comments', function (Property $property) { $property->string('title')->required(); $property->string('content')->required(); $property->string('slug')->pattern('^[a-z0-9-]+$')->required(); $property->object('author', function (Property $author) { $author->string('name')->required(); $author->string('email')->required(); $author->string('bio'); })->required(); $property->array('comments')->items('object', function (Property $comment) { $comment->string('text')->required(); $comment->string('author_name')->required(); $comment->string('author_email')->required(); $comment->int('timestamp')->required(); }); $property->array('tags')->items('string')->minItems(1); $property->enum('status', ['draft', 'published', 'archived']) ->default('draft'); $property->int('views')->min(0)->default(0); })->toArray();
LLM Providers
Each provider has its own structured output format and supports a different subset of JSON Schema. Forerunner builds the payload for you and adjusts the schema to fit.
$struct = Struct::define('Invoice', 'An invoice', function (Property $property) { $property->string('number')->required(); $property->float('total')->min(0)->required(); $property->string('notes')->maxLength(500); });
| Method | Use it as | What it changes |
|---|---|---|
toOpenAI() |
Chat Completions response_format |
Every field becomes required; optional fields become nullable; all objects closed |
toOpenAIResponses() |
Responses API text.format |
Same as above |
toAnthropic() |
Claude Messages API output_config.format |
Unsupported constraints (minimum, maxLength, minItems, ...) are moved into the field description; all objects closed |
toAnthropicTool() |
A strict Claude tool definition (tools[]) |
Same as above |
toGemini() |
Gemini generationConfig |
Unenforced constraints (minLength, maxLength, uniqueItems) are moved into the field description |
toSchema() |
Anything else | Nothing: the plain JSON schema |
$schema and $id are left out of provider payloads. Names are made safe for APIs that only accept letters, numbers, _ and -.
OpenAI
$response = OpenAI::chat()->create([ 'model' => 'gpt-5', 'messages' => [['role' => 'user', 'content' => 'Extract the invoice...']], 'response_format' => $struct->toOpenAI(), ]);
Anthropic
$message = $client->messages->create( model: 'claude-opus-5-5', maxTokens: 16000, messages: [['role' => 'user', 'content' => 'Extract the invoice...']], outputConfig: ['format' => $struct->toAnthropic()], );
Gemini
Http::withHeaders(['x-goog-api-key' => $key])->post($url, [ 'contents' => [['parts' => [['text' => 'Extract the invoice...']]]], 'generationConfig' => $struct->toGemini(), ]);
Choosing a provider at runtime
use Blaspsoft\Forerunner\Providers\Provider; $struct->for(Provider::Anthropic); $struct->for(config('ai.provider')); // 'openai', 'anthropic' or 'gemini'
Working with Generated Schemas
The Struct::define() method returns a Struct object that can be converted to an array or JSON.
Convert to Array
$struct = Struct::define('User', 'A user schema', function (Property $property) { $property->string('name')->required(); $property->string('email')->required(); }); // Convert to array $array = $struct->toArray();
JSON Serialization
The Struct object implements JsonSerializable, so you can use it directly with json_encode():
$struct = Struct::define('User', 'A user schema', function (Property $property) { $property->string('name')->required(); }); // Automatic JSON serialization $json = json_encode($struct, JSON_PRETTY_PRINT);
Using Structure Classes
When using the make:struct command, you can create reusable schema classes:
// In your structure class (generated by make:struct) class UserProfile { public static function schema(): Struct { return Struct::define('user_profile', 'A user profile schema', function (Property $property) { $property->string('name')->required(); $property->string('email')->required(); }); } } // Using the structure UserProfile::schema()->toSchema(); // Plain JSON schema array UserProfile::schema()->toAnthropic(); // Or any provider payload // For JSON, use json_encode $json = json_encode(UserProfile::schema(), JSON_PRETTY_PRINT);
Testing
Run the test suite:
composer test
Run tests with coverage:
composer test-coverage
Code Quality
Run PHPStan analysis:
composer analyse
Format code with Laravel Pint:
composer format
Changelog
Please see CHANGELOG for more information on what has changed recently.
Contributing
Please see CONTRIBUTING for details.
Security Vulnerabilities
Please review our security policy on how to report security vulnerabilities.
Credits
License
The MIT License (MIT). Please see License File for more information.