DocWire SDK Files

This release introduces a major update to the integrated OpenAI functionalities, adding support for the latest models like GPT-4o and retiring older ones. It also enhances dependency management by replacing a custom library port with a standard one, and improves the intelligence of several AI-powered features through better prompting.

The old models fade, a starlit past,
New constellations, brighter and vast.
With prompts refined and logic so keen,
A cleaner build, a powerful scene.
✨🤖🔧

• Features
• Expanded OpenAI Model Support: Integrated support for the latest generation of OpenAI models, including gpt-4o, gpt-4o-mini, gpt-4.1, and the o3 family. This provides access to state-of-the-art AI capabilities for all relevant features.

• Granular Model Selection for Transcription and TTS:

  • Users can now select specific models for audio transcription, with new options like gpt-4o-transcribe and gpt-4o-mini-transcribe, in addition to whisper-1.
  • Added new Text-to-Speech models, including gpt-4o-mini-tts, and set it as the new default for higher quality voice generation.

• Improvements

• Modernized Dependency Management: Replaced the custom unzip vcpkg port with the standard minizip port, simplifying the build process and improving maintainability.
• Enhanced AI Prompting: Significantly improved the system prompts for the Classify and Find features to yield more precise and consistently formatted results from the language models.
• Updated Default Model: The default model for most OpenAI operations has been updated to gpt-4o to leverage its improved intelligence, speed, and cost-effectiveness.

• Robust Example Tests: Updated examples in the documentation to use fuzzy string matching for assertions, making tests more resilient to minor variations in AI-generated text.

• Refactor

• API Modernization: Removed support for deprecated and legacy OpenAI models (e.g., gpt-3.5-turbo, gpt-4-turbo-preview) to streamline the API and focus on current-generation models.
• Transcription Component: Refactored the openai::Transcribe component to allow model selection, making it more flexible and future-proof.