v0.2.0 — Document Indexing & Text Format Support

What's New

Document Indexing

• PDF extraction via PyMuPDF4LLM with async support
• Markdown header-based chunking with hierarchy tracking
• YAML/JSON structural key-path chunking
• HTML/XML tag stripping and plaintext chunking
• Plaintext formats (.txt, .rst, .csv, .log, .ini, .cfg, .toml, config files, dotfiles)

Drift Adapter

• Orthogonal Procrustes transformation for embedding model migration without re-indexing
• Train on corpus sample, save/load rotation matrix

Per-Project Ignore Config

• .tesseraignore with .gitignore syntax (via pathspec)
• Two-tier system: security-critical patterns (.env*, *.pem, etc.) are un-negatable
• User-overridable default patterns

Search Enhancements

• source_type filter on search() — query code, docs, or both
• doc_search() convenience tool for document-only queries
• SQL-level source_type filtering in keyword search for accurate results on skewed corpora
• Over-fetch + post-filter for semantic search with source_type constraints

Infrastructure

• Schema migration system (_meta table, versioned migrations)
• 5 new document columns on chunk_meta (source_type, section_heading, key_path, page_number, parent_section)
• doc_search_tool and drift_train MCP tools

Stats

• 447 tests passing
• ~1,500 LOC production code added
• ~400 LOC tests added
• 4 new dependencies: pymupdf4llm, pathspec, pyyaml, scipy