CHANGELOG.md

[0.4.42] - 2025-08-28

Fixed

• Complete fix for Gemini video validation - consistent field naming (PR #37)
  • Fixed inconsistent camelCase/snake_case in file path video processing
  • File path inputs now use inline_data and mime_type (snake_case)
  • Ensures consistency with base64 input handling from v0.4.41
  • Fully resolves Issue #35: Video input fails with Gemini

[0.4.41] - 2025-08-28

Fixed

• Fixed Gemini video content validation error (PR #36)
  • Video content now properly wrapped in types.Part objects with inline_data
  • Prevents "Extra inputs are not permitted" Pydantic validation errors
  • Maintains full compatibility with base64 and file path video inputs
  • Preserves existing MIME type detection logic

0.4.4 - Gemini 2.5 Flash Image support

• Add text-to-image, image edit (image+prompt), and multi-image fusion via unified CompletionClient
• Model availability checks and graceful skips for tests/examples
• Non-breaking response: text in choices[0].message.content; images exposed in response["images"]
• Examples and tests added; README updated

Changelog

[0.4.3] - 2025-08-21

Added

• Universal system_instruction abstraction across all providers
  • Automatically converts to provider-specific format (system message for OpenAI/Ollama, parameter for Gemini)
  • Seamless cross-provider compatibility
  • No code changes needed - same parameter works everywhere

Changed

• CompletionClient now handles system_instruction conversion internally
• OpenAI provider simplified (no longer needs parameter filtering)

[0.4.2] - 2025-08-21

Fixed

• Fixed OpenAI provider error with 'system_instruction' parameter (Issue #32)
  • OpenAI provider now properly filters out Gemini-specific parameters
  • Added parameter exclusion for 'system_instruction' and 'safety_settings'
  • Ensures cross-provider compatibility when using unified CompletionClient

[0.4.1] - 2025-08-19

Changed

• BREAKING: Image generation now requires explicit model selection
• Renamed GPT4oImageProvider to OpenAIImageProvider for clarity
• Added support for GPT-Image-1 model (available now, not April 2025)
• Removed automatic model fallback - each request must specify the model

Added

• ImageModel enum for explicit model selection (DALLE_2, DALLE_3, GPT_IMAGE_1)
• Model-specific parameter validation
• Token usage tracking for GPT-Image-1
• Support for GPT-Image-1 4K resolution (4096x4096)

Fixed

• GPT-Image-1 response_format parameter handling
• Model-specific size validation
• Proper error messages for invalid models

[0.4.0] - 2025-08-19

Added

• Image Generation Support: New GPT4oImageProvider for DALL-E 3 image generation
• Support for multiple image sizes (SQUARE, LANDSCAPE, PORTRAIT)
• Image quality options (STANDARD, HD)
• Style options (VIVID, NATURAL)
• Image editing capabilities with optional masks
• Azure OpenAI image generation support
• Localized image generation for app internationalization
• C2PA metadata extraction for AI-generated content authenticity
• Comprehensive test suite for image generation

Changed

• Updated package structure to include image providers
• Enhanced README with image generation examples

Technical

• Abstract provider pattern for future image generation services
• Integration with existing BaseCompletions architecture
• Full compatibility with uv package manager

[0.3.9] - 2025-08-09

• Version bump for PyPI release

[0.3.8] - 2025-08-09

• Add /api/models endpoint for listing available models
• Fix GoogleCompletions to query actual Google Gemini API for model list (60+ models!)
• Make list_models() async to properly query provider APIs
• Fix ModelManager fallback logic when no providers are registered
• Improve CompletionClient to handle both sync and async list_models methods
• Update .env.example with comprehensive configuration options
• Add support for filtering models by provider (/api/models?provider=gemini)

[0.3.7] - 2025-08-08

• Restructure project: Move ai_proxy_core source to src/ directory
• Rename api_demo to api_layer for clarity
• Update documentation to use uv run instead of python
• Remove backup file gemini_live_backup.py.old
• Organize test files into tests/ directory
• Create setup/ directory for build artifacts
• Maintain package name as ai_proxy_core for PyPI compatibility

[0.3.6] - 2025-08-08

• Remove legacy completions.py file (replaced by completion_client.py)
• Remove deprecated CompletionsHandler export
• Rename api folder to api_layer to clarify it's the API layer
• Clean up redundant code while maintaining full functionality
• Verified Gemini works with modern CompletionClient

[0.3.5] - 2025-08-08

• Clean up WebSocket file redundancies
• Remove duplicate gemini_live_fixed.py file
• Archive backup WebSocket implementation
• Organize debug and test files into debug/ folder
• Maintain full WebSocket functionality while reducing code duplication
• Tested and verified all endpoints working correctly

[0.3.4] - 2025-08-08

• Fix GoogleCompletions message format handling for google-genai library
• Convert complex message formats to simple string prompts for compatibility
• Fix message parsing to handle both string and list content types
• Improve error handling for API responses
• Ensure proper text extraction from structured message content

[0.3.3] - 2025-08-04

• Fix Gemini Live WebSocket implementation
• Implement proper async context manager handling for sessions
• Add bidirectional communication with separate async tasks
• Fix message format compatibility between client and server
• Add HTML demos for Gemini Live WebSocket (examples/gemini_live_demo.html)
• Add debug tools for testing WebSocket connections
• Reorganize test files into proper directories
• Note: Audio input pending (requires PCM format conversion from WebM)

[0.3.2] - 2025-08-04

• Add security scaffolding for future implementation
• Add optional use_secure_storage parameter to providers (GoogleCompletions, OpenAICompletions)
• Add security dependencies to setup.py extras ([security], [vault], [aws])
• Add TODO placeholders for authentication and secure key storage implementation
• Update CompletionClient to support secure storage configuration
• Maintain full backward compatibility - no breaking changes
• Related to issues #19 and #20

[0.3.1] - 2025-08-03

• Fix telemetry NoOpTelemetry missing request_counter attribute
• Resolve GitHub issues #15, #16, #17
• Add LLM prompt reference for easy integration help
• Add AI Proxy Core expert agent for Claude
• Add roadmap link to README

[0.3.0] - 2025-08-03

• Introduce CompletionClient unified interface
• Add model management system with ModelManager
• Implement model discovery and automatic provider detection

[0.2.0] - 2025-08-03

• Major refactor to provider architecture
• Add support for multiple LLM providers simultaneously

[0.1.9.0] - 2025-08-03

• API migration to provider architecture completed

[0.1.9] - 2025-08-02

• BREAKING: Refactor to provider-specific classes (GoogleCompletions, OpenAICompletions, OllamaCompletions)
• Add multi-provider support with clean separation
• Add optional OpenTelemetry support for basic observability
• Track request counts by model/status across all providers
• Track session duration for WebSocket connections
• Add OpenAI-compatible endpoint support (Groq, Anyscale, etc.)
• Deprecate CompletionsHandler in favor of provider-specific classes
• Add optional dependencies for different providers

[0.1.8] - 2025-08-02

• Add built-in tool support (code_execution, google_search) for Gemini Live
• Add custom_tools parameter for user-defined function declarations
• Update README with comprehensive tool examples
• Add release.sh script for automated releases

[0.1.7] - 2025-08-02

• (Previous release - changes not documented)

[0.1.6] - 2025-01-31

• Add system instruction support for Gemini Live
• Improve WebSocket response handling
• Add TEXT modality to default config

[0.1.5] - 2025-01-30

• Refactor to Python package structure
• Add ai_proxy_core module

[0.1.0] - 2025-01-29

• Initial release
• Basic Gemini completions handler
• Gemini Live WebSocket support