New Features
- **Multiple AI Provider Integrations**
- Added `AnthropicProtector` for Claude models
- Added `CohereProtector` for Cohere models
- Added `DeepSeekProtector` for DeepSeek models
- Added `OpenRouterProtector` for accessing multiple LLMs through OpenRouter
- Restructured `OpenAIProtector` as part of the providers system
- **Enhanced Filtering Patterns**
- Created a dedicated `filtering_patterns` package for better organization
- Implemented advanced LLM injection detection patterns
- Added PII (Personally Identifiable Information) detection
- Developed doxxing prevention capabilities
- Implemented toxicity detection and content moderation
- Added support for custom patterns through `CustomPatternManager`
- **Improved Context Management**
- Redesigned the context manager with `ContextManagerBase`
- Added `TokenBasedContextManager` for token-level context tracking
- Added `MessageBasedContextManager` for conversation contexts
- Added `ContextWindowManager` for dynamic context window handling
- Enhanced `TextCleaner` with more sanitization options
- **Tokenizer Protection**
- Redesigned tokenizer protection with `ReskProtectorTokenizer`
- Added `ReskWordsLists` for efficient prohibited content management
- Improved encoding/decoding protection mechanisms
- **Framework Integration**
- Added Flask integration for securing web APIs
- Added LangChain integration for workflow security
- Added LangGraph integration for agent graphs
- Added Hugging Face integration for model security
- **Autonomous Agent Security**
- Implemented `AgentIdentityManager` for authentication
- Added `AgentSecurityMonitor` for activity tracking
- Developed `AgentSandbox` for confined execution
- Created `SecureAvatar` for user interaction
- **Deployment and Testing**
- Added comprehensive deployment tests
- Improved test coverage
- Added integration tests for all providers
Improvements
- Restructured the codebase for better maintainability
- Enhanced documentation with academic references
- Updated `README.md` with comprehensive examples
- Added example scripts in the `examples` directory
- Updated dependencies in `setup.py` and `requirements.txt`
- Improved error handling and logging
Changed
- Renamed `SecureTokenizer` to `ReskProtectorTokenizer`
- Updated library version to 0.3.0
- Changed import structure in `__init__.py`