README

🌱 AiSeeder

AI-powered database seeding for Laravel 12 — realistic, schema-aware, zero-hallucination.

Stop writing fake data by hand. Stop trusting the AI with your IDs.
AiSeeder reads your database schema, resolves every relationship, and lets the AI generate only the creative parts — while PHP handles everything that must be exact.

⚡ Why AiSeeder?

Traditional seeders require you to hand-craft factories for every table. Faker gives you random strings — not contextually realistic data. And if you ask a raw LLM to generate INSERT statements, it will hallucinate IDs, violate foreign keys, and break JSON columns.

AiSeeder solves all of this:

Problem	AiSeeder's Solution
AI invents fake IDs / ULIDs	PHP generates all PKs & FKs — AI never sees them
Parent table is empty	Recursively seeds parent tables automatically
`VARCHAR(2)` gets `"English"`	Schema analyzer extracts max lengths & ENUMs
JSON columns get plain strings	Strict structured output + post-processing
`password` column gets `"abc123"`	Auto-detected & filled with `Hash::make()`
No visual feedback during generation	Beautiful CLI with spinners, progress bars & token tracking

📦 Installation

composer require mrshennawy/ai-seeder

The service provider is auto-discovered. Publish the config to customize defaults:

php artisan vendor:publish --tag=ai-seeder-config

Make sure you have at least one AI provider configured in your config/ai.php:

# Any provider supported by the Laravel 12 AI SDK
OPENAI_API_KEY=sk-...
# or ANTHROPIC_API_KEY=sk-ant-...
# or GEMINI_API_KEY=...
# or a local Ollama instance

🚀 Quick Start

# Seed 10 rows into the users table (default)
php artisan ai:seed users

# Seed 100 rows
php artisan ai:seed orders --count=100

# Fully interactive mode — pick a table, count, and language
php artisan ai:seed

That's it. AiSeeder will analyze your users table, detect that id is a ULID, password needs hashing, and email must be unique — then ask the AI to generate only the creative columns like name, bio, and phone.

🔍 Feature Deep Dive

1. Smart Schema Introspection

AiSeeder reads your database schema at runtime — not your migrations, not your models — the actual database state. It extracts:

✅ Column names & data types
✅ VARCHAR(n) max lengths — enforced as hard limits on the AI and truncated in PHP as a safety net
✅ ENUM('active','inactive','pending') — AI is constrained to only these values
✅ UNIQUE constraints — AI generates distinct values per row
✅ NULLABLE columns — AI is instructed to occasionally return null
✅ JSON / JSONB columns — AI returns structured objects/arrays, PHP json_encode()s them
✅ Password columns (password, password_hash, etc.) — auto-detected by name

 📋 Table: users — 10 row(s) to generate

 ┌──────────────────┬───────────┬─────────────────────────────┐
 │ Column           │ Type      │ Flags                       │
 ├──────────────────┼───────────┼─────────────────────────────┤
 │ id               │ char      │ PK (ULID), LEN(26)          │
 │ name             │ varchar   │ LEN(255)                    │
 │ email            │ varchar   │ UNIQUE, LEN(255)            │
 │ password         │ varchar   │ PASSWORD, LEN(255)          │
 │ language         │ varchar   │ LEN(2)                      │
 │ status           │ enum      │ ENUM(active|inactive)       │
 │ bio              │ text      │ NULL                        │
 │ preferences      │ json      │ NULL, JSON                  │
 │ created_at       │ timestamp │ NULL                        │
 │ updated_at       │ timestamp │ NULL                        │
 └──────────────────┴───────────┴─────────────────────────────┘

From this analysis, AiSeeder excludes id, password, created_at, and updated_at from the AI prompt entirely. The AI only generates: name, email, language, status, bio, and preferences.

2. Bulletproof IDs & Foreign Keys — Zero Hallucination

This is the architectural philosophy that makes AiSeeder production-safe:

The AI generates zero IDs. PHP handles 100% of structural integrity.

Column Type	Who Handles It	How
Auto-increment PK	Database	Excluded from AI prompt entirely
ULID primary key	PHP	`Str::ulid()` injected per row
UUID primary key	PHP	`Str::uuid()` injected per row
Foreign keys (`user_id`, etc.)	PHP	Random valid parent ID via `array_rand()`
`password`	PHP	`Hash::make('password')` injected
`created_at` / `updated_at`	PHP	`now()` injected
`deleted_at`	PHP	Set to `null` (soft-delete default)

The AI never even sees these columns in the prompt. This means:

✅ 0% chance of SQLSTATE integrity constraint violations from hallucinated IDs
✅ 0% chance of invalid ULID format (like "A1" or "1234567890")
✅ 0% chance of FK pointing to a non-existent parent record

3. Auto-Resolving Relationships (Recursive Seeding)

When you seed a child table, AiSeeder checks every foreign key:

Parent has records? → Fetches existing IDs automatically.
Parent is empty? → Pauses, recursively seeds the parent table first, then resumes.

php artisan ai:seed cart_items --count=20 --lang=ar

 🔗 Resolving relationships...

 🔍 Resolving: user_id → users.id
 ⚠️  Parent table [users] is empty. Recursively seeding it first...

     ┌──────────┐
     │ AiSeeder │  ← Recursive child command for [users]
     └──────────┘
     🔍 Analyzing schema for table: [users]...
     🧠 Generating chunk 1/1 (5 rows)...
     ✅ Successfully seeded [users] with 5 rows.

 ✅ Fetched 5 ID(s) from [users].

 🔍 Resolving: cart_id → carts.id
 ✅ Parent table [carts] already has data. Fetched 12 existing IDs.

 🔍 Resolving: product_id → products.id
 ⚠️  Parent table [products] is empty. Recursively seeding it first...
 ...

The entire dependency tree resolves automatically. Language selection (--lang) propagates to all recursive child commands.

🔄 Self-Referencing Tables

Tables like categories with a parent_id → categories.id are handled gracefully:

php artisan ai:seed categories --count=10

 🔍 Resolving: parent_id → categories.id
 🔄 Self-referencing FK [parent_id] on [categories] — table is empty, will use NULL.

AiSeeder detects the self-reference, skips recursive seeding (which would cause an infinite loop), and sets parent_id = NULL for the initial batch — creating root-level categories. If you run it again, subsequent rows will randomly pick from the existing category IDs.

4. Source Code Context Injection — The Power Feature

Database schemas tell you what a column is. But they can't tell you what a content JSON column should look like when delivery_mode = 'online' vs 'in_person'.

AiSeeder can read your actual PHP code:

php artisan ai:seed course_items \
  --context="Modules\Course\Http\Requests\CourseItemRequest" \
  --count=20

 📄 Loading code context from: Modules\Course\Http\Requests\CourseItemRequest
   ✓ Loaded 87 lines of source code for AI context.

AiSeeder uses PHP's ReflectionClass to locate the source file, reads its raw content with file_get_contents(), and injects it directly into the AI prompt. For example, given a FormRequest like:

class CourseItemRequest extends FormRequest
{
    public function rules(): array
    {
        return [
            'type' => 'required|in:lesson,quiz,assignment',
            'content' => 'required|array',
            'content.location' => 'required_if:delivery_mode,in_person',
            'content.meeting_url' => 'required_if:delivery_mode,online',
            'content.duration_minutes' => 'required|integer|min:15',
        ];
    }
}

The AI will generate content as a proper JSON object:

{
  "location": "Riyadh, Building 4, Room 201",
  "duration_minutes": 60
}

Instead of the broken flat array ["location", "Riyadh", "duration_minutes", 60] it would normally produce.

Use cases for --context:

Pass this class...	The AI learns about...
`StoreOrderRequest`	Conditional validation rules, required-with dependencies
`CourseItem` (Model)	`$casts`, morph maps, `$appends`, accessor logic
`PaymentObserver`	Business rules triggered on create
`UserPolicy`	Role/permission constraints for realistic role distribution

💡 Tip: This is especially powerful for e-learning platforms, ERP systems, or any domain where JSON columns carry complex, context-dependent structures — like Quran academy platforms with lesson content varying by delivery_mode.

5. Multi-Language Data Generation

Generate data in any language — or a realistic mix of multiple languages:

# All text in Arabic
php artisan ai:seed users --lang=ar

# Bilingual Arabic + English (simulates a real bilingual platform)
php artisan ai:seed courses --lang=ar,en

# Trilingual dataset
php artisan ai:seed products --lang=es,pt,fr

# Interactive selection (if --lang is omitted)
🌐 What language(s) should the generated data be in? (comma-separated for multiple, e.g., ar,en)

The language instruction is injected at the system prompt level of the AI agent, ensuring authentic names, titles, and descriptions:

# --lang=ar generates:
┌────────────────────────┬───────────────────────────────┐
│ name                   │ email                         │
├────────────────────────┼───────────────────────────────┤
│ محمود الشناوي            │ mahmoud.shennawy@example.com  │
│ فاطمة أحمد              │ fatima.ahmed@example.com      │
│ عبدالله خالد             │ abdullah.khaled@example.com   │
└────────────────────────┴───────────────────────────────┘

# --lang=ar,en generates a mix:
│ سارة محمد               │ sarah.m@example.com           │
│ John Mitchell           │ john.mitchell@example.com     │
│ أحمد يوسف               │ ahmed.youssef@example.com     │

Technical values (emails, URLs, timestamps, IDs) always remain in ASCII/Latin characters.

6. Beautiful CLI UX & Token Tracking

Built entirely with laravel/prompts for a modern, interactive experience:

 ┌ AiSeeder — Smart Database Seeder ────────────────────────────────┐
 │                                                                   │
 │  🔍 Analyzing schema for table: [orders]...                      │
 │  ⏳ Reading columns, indexes, and constraints...                  │
 │                                                                   │
 │  📋 Table: orders — 50 row(s) to generate                        │
 │  ┌────────────┬──────────┬───────────────────────┐               │
 │  │ Column     │ Type     │ Flags                 │               │
 │  ├────────────┼──────────┼───────────────────────┤               │
 │  │ id         │ char     │ PK (ULID), LEN(26)    │               │
 │  │ ...        │ ...      │ ...                   │               │
 │  └────────────┴──────────┴───────────────────────┘               │
 │                                                                   │
 │  🔗 Resolving relationships...                                   │
 │  🔍 Resolving: user_id → users.id                                │
 │    ✅ Fetched 25 ID(s) from [users].                              │
 │                                                                   │
 │  ⚙️  Plan: 50 rows in 1 chunk(s). Language: AR,EN.               │
 │  ◆ Proceed with seeding [orders]? (Yes)                          │
 │                                                                   │
 │  🧠 Generating chunk 1/1 (50 rows)...                            │
 │  ⏳ Waiting for AI to generate data (this may take a moment)...   │
 │    ✓ AI returned 50 row(s). Tokens: 1,847 prompt + 3,291 comp.  │
 │                                                                   │
 │  💾 Inserting chunk 1/1 into [orders] ████████████████████ 100%   │
 │                                                                   │
 │  ✅ Successfully seeded [orders] with 50 rows.                    │
 │                                                                   │
 │  📊 Token Usage Summary                                          │
 │  ┌─────────────────────┬────────┐                                │
 │  │ Metric              │ Tokens │                                │
 │  ├─────────────────────┼────────┤                                │
 │  │ Prompt tokens       │ 1,847  │                                │
 │  │ Completion tokens   │ 3,291  │                                │
 │  │ Total tokens        │ 5,138  │                                │
 │  └─────────────────────┴────────┘                                │
 └───────────────────────────────────────────────────────────────────┘

Token usage is aggregated across all chunks and recursive parent seeding calls, so you always know the full cost of a seeding operation.

📋 Command Reference

php artisan ai:seed [table] [options]

Argument / Option	Description	Default
`table`	The database table to seed (interactive picker if omitted)	—
`--count=N`	Number of rows to generate	`10`
`--chunk=N`	Rows per AI request (smaller = safer for token limits)	`50`
`--lang=CODE`	Language(s) — single (`ar`) or comma-separated (`ar,en,fr`)	`en`
`--context=CLASS`	Fully-qualified PHP class for business logic context	—
`--no-interaction`	Skip all prompts (use defaults)	—

Examples

# Basic usage
php artisan ai:seed users --count=50

# Large dataset with small chunks to avoid token limits
php artisan ai:seed products --count=1000 --chunk=25

# Arabic-only data with FormRequest context
php artisan ai:seed course_items --count=30 --lang=ar \
  --context="Modules\Course\Http\Requests\CourseItemRequest"

# Fully non-interactive (CI/CD, scripts)
php artisan ai:seed users --count=5 --lang=en --no-interaction

⚙️ Configuration

// config/ai-seeder.php

return [
    // Rows per AI request. Smaller = safer for tokens, larger = fewer API calls.
    'chunk_size' => env('AI_SEEDER_CHUNK_SIZE', 50),

    // Default row count when --count is not provided.
    'default_count' => env('AI_SEEDER_DEFAULT_COUNT', 10),

    // Retry attempts on AI failure (malformed JSON, wrong row count).
    'max_retries' => env('AI_SEEDER_MAX_RETRIES', 3),

    // Default language for generated text content.
    'default_language' => env('AI_SEEDER_DEFAULT_LANGUAGE', 'en'),
];

🏗️ Architecture

packages/shennawy/ai-seeder/
├── config/
│   └── ai-seeder.php                  # Published configuration
├── src/
│   ├── Agents/
│   │   └── SeederAgent.php            # Laravel AI SDK Agent (structured output)
│   ├── Console/Commands/
│   │   └── AiSeedCommand.php          # The ai:seed Artisan command
│   ├── Contracts/
│   │   ├── DataGeneratorInterface.php
│   │   ├── RelationshipResolverInterface.php
│   │   └── SchemaAnalyzerInterface.php
│   ├── AiSeederOrchestrator.php       # Programmatic API (non-CLI usage)
│   ├── AiSeederServiceProvider.php    # Auto-discovered service provider
│   ├── ContextExtractor.php           # ReflectionClass-based source reader
│   ├── DataGenerator.php              # Prompt builder + post-processor
│   ├── GenerationResult.php           # DTO: rows + token usage
│   ├── RelationshipResolver.php       # FK resolution + recursive seeding
│   ├── SchemaAnalyzer.php             # Database introspection engine
│   └── TokenUsageTracker.php          # Aggregates tokens across calls
└── tests/
    └── Feature/                       # 90+ Pest tests

All core services are bound via interfaces in the service provider, making them fully swappable and testable.

🌐 Cross-Provider Compatibility

AiSeeder works with any provider supported by the Laravel 12 AI SDK:

Provider	Status	Notes
OpenAI (GPT-4o, etc.)	✅ Fully supported	Best structured output compliance
Google Gemini	✅ Fully supported	Schema sanitized for strict OpenAPI validation
Anthropic Claude	✅ Fully supported	—
Ollama (local)	✅ Fully supported	Great for development without API costs

The structured output schema is carefully built to avoid provider-specific pitfalls:

No additionalProperties key (Gemini rejects it)
No array-type for nullable fields (Gemini rejects ["string", "null"])
Nullability communicated via descriptions instead

🧪 Testing

The package ships with 90+ tests covering schema analysis, post-processing, prompt building, relationship resolution, self-referencing FKs, and Gemini compatibility:

# Run from the main Laravel project
php artisan test packages/shennawy/ai-seeder/tests/

# Or with a filter
php artisan test --filter="self-referencing" packages/shennawy/ai-seeder/tests/

🛠️ Local Development

To use this package from a local path during development:

// composer.json (main Laravel project)
{
    "repositories": [
        {
            "type": "path",
            "url": "./packages/shennawy/ai-seeder"
        }
    ]
}

composer require mrshennawy/ai-seeder:@dev

📄 License

AiSeeder is open-source software licensed under the MIT License.

Built with ❤️ for the Laravel community by Mahmoud Shennawy

mrshennawy / ai-seeder

Maintainers

Package info

Statistics

Security