Extractor Laravel Package

Added

• Configurable default model via config('extractor.model') / EXTRACTOR_MODEL env var (defaults to gpt-4o-mini)
• System prompt support — extractions can include a system message for better JSON output quality
  • Configure globally via config('extractor.system_prompt') / EXTRACTOR_SYSTEM_PROMPT env var
  • Override per-extractor via systemPrompt() method
  • Disabled by default (set EXTRACTOR_SYSTEM_PROMPT to enable)
• New model constants for current OpenAI models:
  • GPT-5 family: GPT_5_4, GPT_5_4_PRO, GPT_5_3, GPT_5_2, GPT_5_2_PRO, GPT_5_1, GPT_5, GPT_5_PRO, GPT_5_MINI, GPT_5_NANO (with automatic parameter handling — uses max_completion_tokens, omits temperature)
  • GPT-4.1 family: GPT_4_1, GPT_4_1_MINI, GPT_4_1_NANO
  • O-series: O4_MINI, O3, O3_MINI, O3_PRO, O1, O1_PRO (with automatic parameter handling for reasoning models)
  • Renamed: GPT_4O, GPT_4O_MINI (cleaner naming)
• ImageContent detail parameter for OpenAI vision API (low, high, auto)
• Laravel 13 compatibility
• JSON response_format now sent with all requests for more reliable structured output
• Added UPGRADING.md migration guide for breaking changes
• Added comprehensive tests for HasValidation, HasDto, Contacts extractor, system prompts

Changed

• BREAKING: Default model changed from gpt-3.5-turbo-1106 to gpt-4o-mini
• BREAKING: $model, $maxTokens, and $temperature parameters in ExtractorManager methods are now nullable with config/extractor fallback chains
• BREAKING: Removed legacy OpenAI Completions API support — all requests now use Chat Completions API
• BREAKING: Engine::extractResponseText() now accepts only ChatResponse and returns string
• Engine simplified to single chat-only codepath with input-driven vision support
• Widened openai-php/laravel constraint from ^v0.10.2|^0.11 to >=0.10.2 <1.0 for broader compatibility
• Feature tests updated to use current model constants

Removed

• BREAKING: Removed deprecated/shutdown model constants: TEXT_DAVINCI_002, TEXT_DAVINCI_003, GPT_3_TURBO_INSTRUCT, GPT_3_TURBO, GPT_3_TURBO_16K, GPT_3_TURBO_1106, GPT_4, GPT4_32K, GPT_4_1106_PREVIEW, GPT_4_VISION, GPT_O1_MINI, GPT_O1_PREVIEW
• Removed isCompletionModel(), isJsonModeCompatibleModel(), isHybridModel(), isOhOne() methods from Engine
• Removed CompletionResponse import and handling from Engine
• Removed unused Contracts\Engine interface

Deprecated

• Engine::GPT_4_OMNI — use Engine::GPT_4O instead (will be removed in v0.6.0)
• Engine::GPT_4_OMNI_MINI — use Engine::GPT_4O_MINI instead (will be removed in v0.6.0)
• Engine::GPT_4o alias — use Engine::GPT_4O instead (will be removed in v0.6.0)

Added

• Added declare(strict_types=1) to all source files for improved type safety
• Added comprehensive test suite with 127+ tests covering Engine, ExtractorManager, Extractors, and Text loaders
• Added Laravel 10/11/12 integration test script (test-laravel-install.sh)
• Added file size validation in Word loader (50MB limit)
• Added proper error messages for vision model input validation
• Added justfile for simplified development commands
• Added CONTRIBUTING.md with contributor guidelines
• Added pint.json for consistent code formatting configuration
• Added support for custom OpenAI API endpoints (Ollama, Together.ai, Groq, Azure)

Changed

• Improved JSON decode error handling with json_last_error_msg() for better debugging
• Fixed Factory::fromMime() rescue fallbacks to use lazy evaluation (closures)
• Fixed loose equality operators (==) to strict equality (===) in ImageContent
• Improved temp file cleanup with try-finally blocks in Word loader
• Updated temp file prefixes from legacy names to extractor_*
• Fixed Extractor::prompt() to explicitly call ->render() on View objects
• Reorganized test structure into Unit/Feature/Integration directories
• Updated CLAUDE.md with comprehensive architecture documentation

Fixed

• Fixed placeholder exception message in Engine.php for vision model errors
• Fixed eager evaluation bug in Factory::fromMime() fallback handling

Added

• Added support for Google Docs exported Word files
• Added new test case for Google Docs compatibility

Changed

• Improved Word document extraction reliability with better fallback handling
• Enhanced Word.php loader implementation with PHPWord integration
• Improved error handling with proper exception catching
• Better resource cleanup after document processing

Added

• Laravel 12 compatibility

Changed

• Bumped dependencies for Laravel 12 support

Contributors

• @laravel-shift

Fixed

• Minor bug fix release

Changed

• Fixed ability to use GPT-4o for vision tasks
• Deprecated the vision preview model (gpt-4-vision-preview) that no longer works
• Updated documentation to reference new model

Contributors

• Thanks to @blorange2 for providing a fix

Changed

• Updated dependencies
• README improvements

Added

• Support for GPT-4o file handling

Contributors

• @andreascreten

Changed

• Bumped smalot/pdfparser from v2.9.0 to ^2.10

Added

• Laravel 11 support

Fixed

• Fixed broken tests

Changed

• Removed unused prinsfrank/standards dependency
• Bumped dependencies:
  • league/flysystem-aws-s3-v3: ^3.16 → ^3.22.0
  • openai-php/laravel: ^0.7.0 → ^v0.8.1
  • smalot/pdfparser: * → v2.8.0
  • spatie/laravel-data: ^3.9 → ^3.11.0
  • spatie/laravel-package-tools: ^1.14.0 → ^1.16.2

Added

• Hook to enable deletion of files after processing with Textract (cleanupFileUsing)
• Hook to customize file path generation (generateFilePathUsing)
• Added Mockery as dev dependency

Fixed

• Corrected default value mix-up between textract_version and textract_secret in config

Added

• Added fromMime($mime, $content) method to Text\Factory for automatic loader selection based on MIME type

Added

• Support for GPT-4-Vision API
• ImageContent class for handling image inputs

Changed

• Bumped symfony/dom-crawler to ^7.0.0

Fixed

• Fixed OpenAI package installation instructions (config publishing)

Changed

• Set a default model for extractions

Changed

• Moved images out of .github folder for better visibility
• Added example in README header
• Reorganized README structure

Fixed

• Removed unused config options (defer to openai package)

Fixed

• Fixed autoloading configuration

Changed

• Specify output key via variable in prompts

Added

• Initial release
• OpenAI Chat and Completion endpoint wrapper
• Support for Plain Text, PDF, RTF, Images, Word documents, and Web content
• Field Extractor for arbitrary data extraction
• Integration with AWS Textract for OCR
• JSON Mode support for GPT-3.5 and GPT-4 models
• Spatie/data DTO integration
• Built-in extractors: Fields, Contacts, Receipt