v0.2.0
PromptTrail v0.2.0 Release Notes
We're excited to announce the release of PromptTrail v0.2.0, featuring significant improvements in modular agent development, model support, and developer experience.
Major Features
Subroutines for Modular Agents
- Introduced
SubroutineTemplatefor creating modular, reusable components within larger agents - Added flexible session initialization with
SessionInitStrategy:CleanSessionStrategyfor isolated contextsInheritSystemStrategyfor system message preservationLastNMessagesStrategyfor recent context retentionFilteredInheritStrategyfor custom message selection
- Implemented
SquashStrategyfor post-execution message handling:LastMessageStrategyfor final output retentionFilterByRoleStrategyfor role-based filteringLLMFilteringStrategyandLLMSummarizingStrategyfor AI-powered context management
Enhanced Model Support
- Added support for Google Gemini and Anthropic Claude Sonnet models
- Added local LLM support using Hugging Face Transformers
API Improvements
- Unified configuration system:
- Merged
Parametersinto consolidatedConfigclass - Streamlined model settings and API key management
- Merged
- Introduced unified
ToolingTemplatesupporting both OpenAI and Anthropic models - Standardized message roles across all models
- Implemented new logging system with
Debuggablebase class
Technical Updates
Dependency Upgrades
- anthropic: v0.45.2
- openai: v1.61.0
- google-generativeai: v0.8.4
Code Quality Improvements
- Renamed
Message.sendertoMessage.rolefor industry standard alignment - Simplified model class names (e.g.,
OpenAIChatCompletionModeltoOpenAIModel) - Enhanced type safety in tool system
- Transitioned to session-based state management
- Implemented template-driven API enhancements
- Streamlined metadata handling in Session class
Developer Tools
- Added automated code review bot (
examples/dogfooding/auto_code_review.py) - Enhanced project documentation and tooling processes
- Improved debugging capabilities through standardized logging
For detailed documentation and migration guides, please refer to our documentation website.