Releases: combinatrix-ai/PromptTrail
Releases · combinatrix-ai/PromptTrail
v0.2.1
-
Feature: Model Override for Templates
- Added the ability to override the model used in
GenerateTemplateandToolingTemplate. - This allows for more flexible and cost-effective model selection within agents.
- Includes design documentation and examples for template model overrides.
- Added the ability to override the model used in
-
Refactor: Enhanced Conversation End Handling
- Removed
JumpTemplatefor simplified conversation flow control. - Introduced
EndConversationTooland farewell messages inEndTemplatefor more robust end-of-conversation handling. - Simplified runner by removing
start_template_id. - Added comprehensive tests and design documentation for end conversation features.
- Removed
-
Feature: Event Handling System
- Implemented an event handling system with
UserInteractionEventto manage user interactions more effectively. - Refactored core components to improve maintainability and clarity.
- Renamed session transformers for better understanding.
- Includes a new design proposal for event handling.
- Implemented an event handling system with
-
Feature: API Runner with FastAPI Integration
- Introduced
APIRunnerwith FastAPI for creating web-based APIs for agents. - Supports session management, RESTful endpoints, asynchronous operations, and session serialization.
- Introduced
-
Module Restructuring and Proposals
- Restructured the core module by moving core implementation to
_core.pyfor better organization. - Added new proposal documents for future features.
- Restructured the core module by moving core implementation to
-
Enhancements to Release Workflows and Documentation
- Implemented an automated release process for smoother updates.
- Improved documentation with details on runners, debugging, and new features.
- Added a weather forecast API example to showcase API runner capabilities.
- Enhanced markdown code block extraction and updated README.md.
- Removed outdated roadmap from README.
v0.2.1rc1
Feature: Model Override and Core Improvements
- Model Override Capability: Introduced model override functionality for
GenerateTemplateandToolingTemplate. This allows for specifying a different model for individual templates, enabling cost and performance optimization. - Enhanced Jinja2 Error Handling: Improved Jinja2 template error handling by adding a
disable_jinjaoption. This is useful when template content includes Jinja syntax that should not be rendered. - Subroutine Environment Isolation: Implemented complete environment isolation for
SubroutineTemplate, ensuring subroutines run in a clean context without affecting the parent session. - Runner Model Handling: Refactored runner model handling for better consistency and clarity across the framework.
- Design Documentation: Added comprehensive design documentation for template model override, detailing its benefits, usage, and implementation.
- Examples Update: Updated examples to showcase new features and improve overall code structure, providing clearer usage patterns.
Refactor: End Conversation Flow
- Removed
JumpTemplate: RemovedJumpTemplateto simplify conversation flow control. Navigation logic should be handled by control flow templates and session state. - Enhanced
EndTemplate: ImprovedEndTemplateto support optional farewell messages, providing a more graceful way to end conversations. - Simplified Runner: Runner implementation is simplified by removing
start_template_id, streamlining the execution process. EndConversationTool: AddedEndConversationToolto allow LLMs to explicitly end conversations, with support for farewell messages.- Design Documentation: Added design documentation for
EndConversationTool, outlining its purpose, usage, and integration within the framework.
Feature: Event Handling and Core Refactoring
- Event Handling System: Implemented a new event handling system, introducing
UserInteractionEventfor managing user inputs and interactions within templates. - Core Component Refactoring: Refactored core components to enhance maintainability and scalability, improving the internal structure of the framework.
- Session Transformer Renaming: Renamed session transformers for improved clarity and consistency in naming conventions.
- Design Proposal: Added a design proposal document for the event handling system, detailing its architecture, benefits, and future extensions.
Feature: API Runner with FastAPI Integration
APIRunner: Introduced a newAPIRunnerclass, providing seamless integration with FastAPI to expose agents as RESTful APIs.- FastAPI Integration: Implemented REST endpoints for session creation, status retrieval, input posting, and session starting, facilitating web-based agent interaction.
- Session Management: Added robust session management capabilities within
APIRunnerto handle multiple concurrent conversations. - Async Support: Ens
v0.2.1rc0
Checking release pipeline...
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.