v0.20.1

GitHub Release Changelog

🚀 v0.20.1 Release - July 29, 2026

✨ Features

• Kimi K3 Model Support: Added Kimi K3 (MoonshotAI's 2.8T MoE flagship model) to all supported providers
  • CrofAI: kimi-k3 with 1M context, native vision, always-on reasoning
  • Fireworks AI: accounts/fireworks/models/kimi-k3
  • Together AI: moonshotai/Kimi-K3
  • All implementations include tool use, thinking, vision, stream, and structured output capabilities

⚡ Performance Improvements

• Reduced App Load Time: Optimized application startup by implementing:
  • Background thread for version checks (non-blocking startup)
  • Lazy initialization of image generation providers
  • Deferred loading of LLM model definitions
  • Centralized model definitions to minimize import overhead
• Image Generation Optimization: Provider SDK imports now deferred until actual use
• Model Registry Efficiency: Only loads providers whose API keys are available in environment

🔧 Technical Changes

• Background Version Check: Version update checks now run in parallel with app setup (~1.2s savings)
• Lazy Provider Loading: Image generation and LLM providers initialize on-demand rather than at startup
• Centralized Model Definitions: All LLM model definitions moved to new model_definitions module
  • Reduces circular dependency risks
  • Only imports Model type without heavy service dependencies
  • Each provider module can be loaded independently
• Enhanced Version Display: Shows current version at startup with update notifications

🐛 Bug Fixes

• Fixed potential performance bottleneck during application initialization
• Improved error handling for update checks with graceful fallbacks

🏗️ Architecture

• New AgentCrew/modules/llm/model_definitions module structure
• Improved separation of concerns between model definitions and service implementations
• Enhanced provider availability checking without full initialization