Skip to content

Latest commit

 

History

History
119 lines (95 loc) · 7.09 KB

CHANGELOG.md

File metadata and controls

119 lines (95 loc) · 7.09 KB

Changelog

All notable changes to the HelpingAI Python SDK will be documented in this file.

[1.2.3] - 2025-11-14

Fixed

  • 📦 Package Structure: Fixed setuptools packaging configuration to properly include all subpackages (HelpingAI.*), resolving ModuleNotFoundError when importing HAI from installed package
  • 🛠️ Import Paths: Corrected import paths in HelpingAI/__init__.py to ensure all client submodules are accessible when importing the main HelpingAI package
  • 🔄 Backward Compatibility: Verified that existing code importing HAI from HelpingAI continues to function correctly after refactoring
  • 🧹 Code Cleanup: Removed redundant imports and streamlined HelpingAI/__init__.py for clarity and maintainability
  • updated documentation to reflect correct import paths and usage examples

[1.2.2] - 2025-05-08

Added

  • 🔄 Multi-Format API Support: Enhanced models API to support both HelpingAI format (array of strings) and v1/models schema format
  • 🔧 Auto-Detection: Automatic detection of response format in list() method for seamless compatibility
  • 🔗 v1 Models Schema Support: New from_v1_models_data() method for explicit v1/models schema handling
  • 📊 Custom Logging System: New HelpingAI/logging.py with structured logging, colored output, and environment configuration (may get replaced with LITPRINTER)
  • 🎨 Enhanced Error Diagnostics: Advanced error parsing with helpful suggestions and context-aware recommendations

Changed

  • 🧹 Major Refactor: Split HelpingAI/client.py into modular submodules: base.py, completions.py, chat.py, and main.py under HelpingAI/client/ for improved maintainability and clarity
  • 🔄 Backward Compatibility: The original client.py now re-exports all main client classes for seamless transition
  • 🧩 Internal Structure: All client logic and classes are now organized in dedicated files, reducing file size and improving code navigation
  • 🔐 Enhanced Security: Models endpoint now requires authentication (auth_required=True)
  • 🧹 Code Cleanup: Removed all hardcoded model names for fully dynamic API-driven model management
  • 📡 Dynamic Model Loading: Models are now fetched entirely from API responses instead of fallback hardcoded lists

Enhanced

  • 🧠 Simplified Think Logic: Simplified complex hide_think parameter logic as it is now internally handled by HelpingAI's backend
  • 🎯 Flexible Data Handling: Model.from_api_data() now accepts both string (HelpingAI format) and dict (v1/models schema) inputs
  • 🔍 Better Error Reporting: Error messages now show actual available models from API responses
  • 🧩 Format Agnostic: Seamless handling of different API response structures without breaking changes
  • 🛠️ Advanced Error Handling: Modular error parsing with helper methods for message extraction, model name detection, and streaming suggestions
  • 📋 Smart Error Messages: Context-aware error enhancement with actionable suggestions based on status codes and request patterns
  • 🌈 Colored Log Output: Custom logging system with color-coded levels, timestamps, and configurable output destinations
  • ⚙️ Environment Configuration: Logging configuration via environment variables (HAI_LOG_LEVEL, HAI_LOG_FILE, HAI_LOG_CONSOLE)

[1.2.0] - 2025-07-20

Added

  • 🔌 MCP Integration: Full Model context Protocol (MCP) support for external tool connections
  • 🖥️ Multiple Transport Types: Support for stdio, SSE, and streamable-http MCP servers
  • 🔄 Automatic Tool Discovery: MCP tools automatically converted to the standard tool calling format
  • 📁 Resource Support: Built-in list_resources and read_resource tools for MCP resources
  • 🔀 Mixed Tools Support: Seamlessly combine MCP servers with regular standard tool definitions
  • ⚡ Process Management: Automatic cleanup of MCP server processes on exit
  • 🔁 Reconnection Logic: Handles server disconnections automatically
  • 🛡️ Graceful Error Handling: Works without MCP package installed with helpful error messages
  • 📦 Optional MCP Dependency: Install with pip install HelpingAI[mcp] for MCP features
  • New MCP integration documentation and examples

Enhanced

  • 🛠️ Extended Tools Compatibility: Enhanced tools framework to support MCP server configurations
  • 🌐 Popular MCP Servers: Ready support for mcp-server-time, mcp-server-fetch, mcp-server-filesystem, and more
  • 🏗️ Backward Compatibility: Fully backward compatible with no breaking changes to existing functionality

[1.1.3] - 2025-07-18

Added

  • 🔧 Tool Calling Framework: New @tools decorator for effortless tool creation
  • 🔄 Direct Tool Execution: New .call() method on HAI client for executing tools without registry manipulation
  • 🤖 Automatic Schema Generation: Type hint-based JSON schema creation with docstring parsing
  • 📝 Smart Documentation: Multi-format docstring parsing (Google, Sphinx, NumPy styles)
  • 🧠 Thread-Safe Tool Registry: Reliable tool management in multi-threaded environments
  • 🔍 Tool Validation: Automatic parameter validation against JSON schema
  • Extended Python Support: Now supports Python 3.7-3.14
  • Streaming Support: Real-time response streaming
  • Advanced Filtering: Hide reasoning blocks with hide_think parameter
  • New comprehensive Tool Calling Guide

Changed

  • 🔄 Universal Compatibility: Seamless integration with existing standard tool definitions
  • Updated Models: Support for latest models (Dhanishtha-2.0-preview, Dhanishtha-2.0-preview-mini)
  • Improved Model Management: Better fallback handling and detailed model descriptions
  • Simplified Tool Execution: Direct tool calling with client.call(tool_name, arguments) syntax
  • Deprecated get_tools_format() in favor of get_tools()
  • Updated documentation to reflect current model names and best practices

Enhanced

  • 🛡️ Enhanced Tool Error Handling: Comprehensive exception types for tool operations
  • Dhanishtha-2.0 Integration: World's first intermediate thinking model with multi-phase reasoning
  • Dhanishtha Models: Advanced reasoning capabilities with transparent thinking processes
  • Function-Calling Friendly Interface: Familiar API design
  • Enhanced Error Handling: Comprehensive exception types

[1.1.2] - 2025-06-15

Added

  • Support for Dhanishtha-2.0-preview model
  • Improved error handling for API requests
  • Enhanced streaming capabilities

Fixed

  • Various bug fixes and performance improvements

[1.1.1] - 2025-05-20

Added

  • Initial support for tool calling
  • Enhanced type hints for better IDE support

Fixed

  • Connection handling for unstable networks
  • Token counting accuracy

[1.1.0] - 2025-04-10

Added

  • Initial public release
  • Support for chat completions
  • Basic streaming functionality
  • Error handling framework

For more details, see the documentation or GitHub repository.