Skip to content

CHANGELOG.md

Latest commit

 

History

History
120 lines (101 loc) · 6.97 KB

CHANGELOG.md

File metadata and controls

120 lines (101 loc) · 6.97 KB

Changelog

[0.5.0] - 2025-12-08

Added

  • Implemented Ollama integration for AI-powered code explanations with GPU acceleration
    • New ollama_llm.py module with OllamaLLM class for communication with local Ollama service
    • Added --ollama flag to enable Ollama-enhanced query responses
    • Added --ollama-model parameter to specify which Ollama model to use (default: llama3)
    • Added --max-tokens parameter to control response length
    • New generate_answer_ollama() function in rag.py for Ollama-based RAG responses
    • Comprehensive documentation in docs/ollama_integration.md
    • Full test suite for Ollama integration in tests/test_ollama.py, tests/test_rag_ollama.py, and tests/test_cli_ollama.py
  • Implemented cmd_query() CLI function for semantic search with support for hybrid search and documentation merging
  • Enhanced autoplan with index-driven section discovery:
    • _discover_key_modules(): Identifies core modules by symbol count
    • _discover_key_classes(): Finds important classes by structural complexity
    • _discover_key_functions(): Locates critical functions via call graph analysis
    • _discover_patterns(): Detects architectural patterns (testing, configuration, API, data layer)
  • Added PlanningRule dataclass for declarative section planning rules
  • Refactored generate_plan() to support both rule-based and index-discovered sections

Changed

  • Autoplan now generates documentation sections dynamically from indexed codebase rather than hardcoded templates
  • generate_plan() accepts optional index_dir parameter to enable index-based discovery
  • Improved code organization with separation of concerns in planning logic
  • README.md updated to reflect index-driven documentation planning and Ollama integration
  • CLI argument parser enhanced with new Ollama-specific options

Fixed

  • Fixed missing cmd_query function that was causing NameError in CLI
  • Added missing imports (json, sqlite3) to cli.py
  • Fixed unpacking error in query result formatting (6 fields instead of 7)

[0.4.0] - 2025-11-15

Added

  • External library documentation indexing (PyPI + local site-packages)
  • New public Python wrappers: Indexer and Searcher for programmatic indexing and querying
  • New semindex.docs package with automated documentation generation
  • Added scripts/gen_docs.py for generating wiki documentation
  • Implemented LocalLLM with automatic model download for offline documentation generation
  • Added OpenAICompatibleLLM for remote LLM integration

Changed

  • Extended store.py schema to manage docs-specific tables and FAISS index
  • cli.py updated to support docs indexing and retrieval
  • Improved error handling for external service failures

[0.3.0] - 2025-10-20

Added

  • Added a pluggable language adapter registry in semindex.languages that powers automatic discovery of supported file types and supports runtime registration of custom adapters.
  • Introduced an optional Tree-sitter powered JavaScript adapter that is registered when tree_sitter_languages is available, expanding multi-language indexing support.
  • Expanded Tree-sitter backed adapters to cover javascript, java, typescript, csharp, cpp, c, go, php, shell, rust, and ruby (12 languages total with extras) and upgraded the JavaScript adapter to emit class/function symbols via the Tree-sitter AST.
  • External library documentation indexing (PyPI + local site-packages). Docs are parsed (HTML/Markdown), normalized, embedded, and stored in dedicated tables (doc_packages, doc_pages, doc_vectors) and a separate FAISS index docs.faiss. CLI: --include-docs for index and query, with --docs-weight to control ranking merge.
  • New public Python wrappers: Indexer (semindex.indexer.Indexer) and Searcher (semindex.search.Searcher) for programmatic indexing and querying, including hybrid search and optional docs merging.
  • New semindex.docs package exposing generate_plan(), graph builders, and Mermaid utilities to power automated documentation.
  • Added scripts/gen_docs.py CLI for generating wiki documentation from graphs, repo statistics, and LLM-authored narratives.
  • Implemented LocalLLM with automatic TinyLlama GGUF download (configurable via SEMINDEX_LLM_* env vars) for offline documentation generation.
  • Added remote_llm.py with OpenAICompatibleLLM + resolve_groq_config() to integrate Groq/OpenAI-compatible endpoints, and surfaced --remote-llm CLI options in gen_docs.py.
  • Added automated planner tests in tests/test_autoplan.py and CLI coverage in tests/test_gen_docs_cli.py.

Changed

  • Incremental indexing now reuses the adapter registry so mixed-language repositories are handled consistently in both fresh and incremental runs.
  • Extended store.py schema and index reset logic to manage docs-specific tables and FAISS index.
  • cli.py updated to optionally index docs after code indexing and to merge doc results at query time.
  • README.md and ROADMAP.md updated with the documentation generator workflow, LLM configuration, and dependency-group guidance.
  • pyproject.toml now credits OpenSource Syndicate as the author and introduces a languages dependency group for uv-based installs.

0.2.0 - 2025-10-06

Added

  • Hybrid search functionality combining dense vector search and keyword search using Elasticsearch
  • Reciprocal Rank Fusion (RRF) to combine search results from multiple sources
  • Semantic-aware chunking using CAST-like algorithm for better context preservation
  • Incremental indexing to only re-process changed files using file hashing
  • --hybrid flag to enable hybrid search in query command
  • --chunking option to select between symbol-based and semantic chunking
  • --similarity-threshold parameter for controlling semantic chunking sensitivity
  • --incremental flag for incremental indexing mode
  • Comprehensive test suite for all new features

Changed

  • Enhanced CLI to support new hybrid search and chunking options
  • Updated indexing process to support both fresh and incremental indexing
  • Improved error handling for Elasticsearch connection failures
  • Refactored chunking module to support both traditional and semantic-aware chunking
  • Added proper hash-based comparison for Symbol class

Fixed

  • Symbol hashability issue that was causing errors in dictionary lookups
  • Various mocking issues in tests to properly isolate functionality

Dependencies

  • Added elasticsearch dependency for keyword search functionality
  • Added sentence-transformers for semantic similarity calculations

0.1.0 - 2025-10-05

Added

  • Initial release of semindex
  • AST-based parsing for Python code
  • Vector search using FAISS
  • CLI interface for indexing and querying
  • Basic chunking by function/class boundaries
  • SQLite for metadata storage