Changelog

All notable changes to this project are documented here.
Format follows Keep a Changelog.

[0.2.0]

EmbeddingModel (ragmill.embeddings) — local ONNX-based sentence embeddings via a quantized MiniLM model, downloaded once and cached offline.
VectorStore (ragmill.vector_store) — SQLite-backed storage with brute-force cosine similarity search, plus filtering by filename, source_file, modified_after, modified_before.
sync_directory (ragmill.sync) — incremental sync between a folder and a VectorStore: skips unchanged files (content-hash based), replaces chunks for changed files, removes chunks for deleted files.
modified_at is now captured during ingestion and threaded through chunk metadata.

Project renamed to RAGMill. The original name, nexus-flow, was already taken on PyPI by an unrelated package; the next candidate, nexusflow, was rejected by PyPI for being confusingly similar to it. Package name is now ragmill, and the import path changed accordingly: import ragmill (previously import nexus_flow). The main class was renamed NexusEngine → RAGEngine to match.

RAGEngine — directory ingestion (.txt, .md, .log, .rst, .pdf via pypdf, .docx via python-docx) and semantic chunking with paragraph/sentence-boundary splitting and configurable overlap.
Optional extras (pdf, docx, all, dev) so the core package has zero hard dependencies.