v2.1

New Features:

1. Observability with Langfuse
   • Integrated Langfuse tracing to track LLM calls, tool usage, and graph execution across the full pipeline.
   • Added dedicated observability.ipynb notebook with a step-by-step guide to set up and use Langfuse with the system.

2. Streaming Response & Agent Transparency
   • Responses now stream token by token directly in the Gradio interface.
   • Agent reasoning is now visible in real time: tool calls (search_child_chunks, retrieve_parent_chunks) and their results are shown as collapsible messages during execution.
   • System nodes (query rewriting, history summarization) surface their output progressively, including clarification requests when the query is unclear.

Bug Fixes:

1. Docker Volume Permission Fix
   • Fixed a PermissionError when clicking "Clear All" with Docker-mounted volumes.
   • Directories are now cleared by removing their contents instead of deleting the root folder, avoiding permission issues with mount points.

v2.0

New Features:

1. Context Compression
   • Compresses the agent's working memory when the context exceeds a configurable token threshold.
   • Prevents redundant tool calls across long retrieval loops. Tunable via BASE_TOKEN_THRESHOLD and TOKEN_GROWTH_FACTOR.

2. Agent Limits & Fallback Response
   • Introduced hard caps on tool invocations (MAX_TOOL_CALLS) and reasoning loop iterations (MAX_ITERATIONS) to ensure bounded execution.
   • When either limit is reached, the agent falls back to a dedicated response node, using all context retrieved so far to generate the best possible answer rather than failing silently.

Improvements:

1. Enhanced Notebook Documentation
   • Expanded and clarified comments for each code block, including references to official documentation.
   • Improves readability and helps users better understand each pipeline component.

v1.9

Improvements:

1. Improved Notebook Readability
   • Updated notebooks with clearer formatting and structure.
   • Enhances user experience and makes it easier to follow code and outputs.

2. Modular Architecture Documentation Enhanced
   • Expanded documentation for the modular architecture.
   • Provides more clarity on how components interact and can be extended.

3. Updated System Prompts and Tool Outputs
   • Refined prompts and outputs to improve accuracy and consistency.
   • Enhances reliability of pipeline responses and results.

4. Updated Dependencies
   • All dependencies upgraded to their latest versions.
   • Ensures compatibility, security, and performance improvements.

5. README and Links Updated
   • Fixed and updated links for notebooks and diagrams.
   • Simplifies navigation and access to resources.

v1.8

New Features:

1. Multi-Agent Map-Reduce in Tutorial Notebook
   • Fully integrated into the notebook for hands-on experimentation.
   • Enables users to run and inspect the multi-agent pipeline directly when handling complex RAG queries.

2. Alternative PDF-to-Markdown Conversion Tools
   • Added multiple tool options for PDF-to-Markdown conversion within the notebook.
   • Improves flexibility and robustness when dealing with different PDF formats and extraction quality.

Improvements:

1. Enhanced Notebook Documentation
   • Expanded and clarified comments for each code block, including references to official documentation.
   • Improves readability and helps users better understand each pipeline component.

2. Troubleshooting Section Added
   • Introduced a dedicated troubleshooting section to address common issues.

v1.7

New Features:

1. Multi-Agent Map-Reduce for RAG Queries
   • Decomposes complex queries into parallel sub-queries to generate more comprehensive and accurate answers.
   • Enhanced parallelization of retrieval and generation steps for more efficient responses.

Improvements:

1. Codebase Refactoring
   • Performed structural cleanup to improve code readability and maintainability.
   • Simplified components to support easier extension and future development.

2. Updated Tutorial Notebook
   • Improved explanations and examples for better learning flow.
   • Added clearer guidance to help users understand and run the pipeline more effectively.

v1.6

New Features:

1. End-to-End Gradio Interface for RAG Pipeline
   • Integrated a fully functional Gradio interface to streamline interaction with the RAG pipeline.
   • Added PDF ingestion via file upload to populate the knowledge base.
   • Implemented document deletion, allowing users to remove unwanted content from the system.

2. Modular Project Structure
   • Refactored the codebase to adopt a modular architecture.
   • Improved maintainability, readability, and scalability, making it easier to extend and integrate new components.

v1.5

New Features:

1. Conversation Memory via Summarization
   • Implemented a memory mechanism that summarizes past interactions to maintain context across the conversation.

2. Human-in-the-Loop Query Clarification
   • Introduced an interactive clarification step that prompts the user when the query is ambiguous or incomplete.
   • Allows the system to gather additional context before proceeding with retrieval and generation.

v1.0

Initial Release:

1. Agentic RAG Pipeline
   • First public release of a minimal Agentic RAG system built with LangGraph.
   • Implements hierarchical indexing with parent/child chunking strategy for precise and context-rich retrieval.
   • Includes hybrid search via Qdrant with dense and sparse embeddings.