Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[Unreleased]

Settings reload method: Added settings.reload_from_env() method to allow reloading configuration from environment variables after module import. Fixes Colab/Jupyter notebook issue where settings were cached before environment variables were set.

Colab tutorial improvements: Updated Step 5 to explicitly call settings.reload_from_env() and print confirmation of vector store path. Ensures vector store is properly detected in notebooks.
Build script error message: Updated orchestrator error message to use python -m honegumi_rag_assistant.build_vector_store instead of outdated scripts/build_vector_store.py path.

Vector store Ax version pinning: src/honegumi_rag_assistant/build_vector_store.py now defaults to cloning Ax v0.4.3 (matching honegumi dependency) instead of latest main branch. This prevents documentation version mismatch where LLM receives v1.x docs but generates v0.4.3 code. New --ax-version parameter allows specifying different versions when needed. Addresses issue raised in PR #3.
Vector store missing condition now properly detected when path is configured but directory doesn’t exist

Build script location: Moved build_vector_store.py from scripts/ to src/honegumi_rag_assistant/ to make it available as a module for pip-installed users. Usage: python -m honegumi_rag_assistant.build_vector_store
Retrieval UX improvements: Simplified retrieval output to show only essential information. In non-debug mode, displays “Planning X parallel retrievals…” followed by success message with context count and timing, or clear error if vector store missing. Removes verbose query details and individual retriever timings from default output.
Vector store metadata: Build script now saves metadata.json with build details including Ax version, build date, chunk configuration, and document counts for better provenance tracking.
Removed verbose per-query retrieval details from non-debug output
Cleaner success message: “✓ Retrieved X contexts in Y.XXs”

Simplified ReadTheDocs navigation structure (removed unnecessary sections)
Reduced documentation table of contents depth from 2 to 1 for cleaner appearance
Removed License and Authors pages from main navigation

Updated project name from “honegumi_rag_assistant” to “Honegumi RAG Assistant” in documentation
Replaced Features section with Key Capabilities from README in docs/index.md
Added project overview and description to documentation landing page
Updated README.md badge to track stable version instead of latest
Improved documentation landing page with logo and better structure

Moved build_vector_store.py from scripts/ to src/honegumi_rag_assistant/ for pip package accessibility
Configured PyPI to use README_PYPI.md for concise package description
ReadTheDocs configuration with Google-style docstring support
Sphinx ReadTheDocs theme
Napoleon extension configured for Google docstrings
Comprehensive documentation structure

Initial release of Honegumi RAG Assistant
Multi-agent LangGraph pipeline for code generation
RAG-based retrieval from Ax Platform documentation
FAISS vector store for fast document retrieval
Specialized agents: IssueScout, ParameterSelector, RetrievalPlanner, CodeWriter, Reviewer
CLI interface with honegumi-rag command
Programmatic API via run_from_text() and run_from_dict()
Debug mode for detailed execution logging
Optional code review and refinement step
Support for custom output directories