Ra-aid

Added
- Support for Fireworks.ai LLM provider with error handling
- Support for Groq provider
- Cloudflare build scripts and logging

Changed
- Updated model parameters and providers configuration
- Multiple package-lock.json updates

Fixed
- npm version specification in package.json

Added
- Custom Tools Feature
- Added support for custom tools with `--custom-tools <path>` CLI flag
- Implemented MCP (Model-Completion-Protocol) client for integrating external tool providers
- Created documentation on custom tools usage in `docs/docs/usage/custom-tools.md`
- Added example code in `examples/custom-tools-mcp/` directory
- API Documentation
- Added comprehensive OpenAPI documentation for REST API endpoints
- Implemented API documentation in Docusaurus with new MDX files
- Added YAML OpenAPI specification file `docs/ra-aid.openapi.yml`
- Created script to generate OpenAPI documentation automatically
- Session Usage Statistics
- Added CLI commands for retrieving usage statistics for all sessions and the latest session
- Enhanced session and trajectory repositories with new methods
- Moved scripts into proper Python package structure (`ra_aid/scripts/`)
- Web UI Improvements
- Added new UI components including input box, session screen, and buttons
- Improved session management UI
- Enhanced styling and layout

Changed
- WebSocket Endpoint Migration
- Migrated WebSocket endpoint from `/ws` to `/v1/ws` to align with REST API endpoint pattern
- Updated root HTML endpoint to reflect the new WebSocket path
- Project Maintenance
- Refactored agent creation logic to use model capabilities for selecting agent type
- Improved model detection and normalization
- Updated dependencies via uv.lock
- Fixed various typos and improved prompts

Added
- Project State Directory Feature
- Added `--project-state-dir` parameter to allow customization of where project data is stored
- Modified database connection, logging, and memory wiping to support custom directories
- Created comprehensive documentation in docs/docs/configuration/project-state.md
- Ollama Integration
- Added support for running models locally via Ollama
- Implemented configuration options including model selection and context window size
- Added documentation for Ollama in docs/docs/configuration/ollama.md
- Updated open-models.md to include Ollama as a supported provider
- Web UI and API Progress (partially implemented)
- Created API endpoints for session management (create, list, retrieve)
- Added trajectory tracking and visualization
- Implemented UI components for session management
- Added server infrastructure for web interface
- Token Usage and Cost Tracking
- Enhanced trajectory tracking with token counting
- Added session-level token usage and cost tracking
- Improved cost calculation and logging

Fixed
- Fixed bug with `process_thinking_content` function by moving it from `agent_utils` to `ra_aid.text.processing` module
- Fixed config parameter handling in research request functions
- Updated development setup instructions in README to use `pip install -e \".[dev]\"` instead of `pip install -r requirements-dev.txt`

Added
- Added support for think tags in models with the new extract_think_tag function
- Enhanced CiaynAgent and expert tool to extract and display thinking content from <think>...</think> tags
- Added model parameters for think tag support
- Added comprehensive testing for think tag functionality
- Added `--show-thoughts` flag to show thoughts of thinking models
- Added `--show-cost` flag to display cost information during agent operations
- Enhanced cost tracking with AnthropicCallbackHandler for monitoring token usage and costs
- Added Session and Trajectory models to track application state and agent actions
- Added comprehensive environment inventory system for collecting and providing system information to agents
- Added repository implementations for Session and Trajectory models
- Added support for reasoning assistance in research phase
- Added new config parameters for managing cost display and reasoning assistance

Changed
- Updated langchain/langgraph deps
- Improved trajectory tracking for better debugging and analysis
- Enhanced prompts throughout the system for better performance
- Improved token management with better handling of thinking tokens in Claude models
- Updated project information inclusion in prompts
- Reorganized agent code with better extraction of core functionality
- Refactored anthropic token limiting for better control over token usage

Fixed
- Fixed binary file detection
- Fixed environment inventory sorting
- Fixed token limiter functionality
- Various test improvements and fixes

Changed
- Replaced thread-local storage with contextvars in agent_context.py for better context isolation
- Improved React agent execution with LangGraph's interrupt mechanism
- Enhanced _run_agent_stream function to properly handle agent state and continuation

Fixed
- Fixed tests to work with the new implementation