VERA — Vector-Embedded Retrieval Archive
A tool that converts documents into single, portable .vera SQLite archive files holding text, embeddings, a keyword index and citations, then supports keyword, semantic or hybrid search over one file or a folder of them, with no database server.
VERA converts a document, such as a PDF or Markdown file, into a self-contained .vera SQLite archive that holds its text, embeddings, keyword index, citations and the original source together, so the file can be copied, shared or handed to an AI agent without running a database server. A folder of converted files plus a built index becomes a searchable library.
Its default embedder is a fast, offline hashing model based on shared words, which supports keyword, semantic and hybrid search modes but does not capture meaning or synonyms; installing a local model runtime and converting with a neural model such as MiniLM enables meaning-based semantic and hybrid search instead. PDFs get automatic OCR when needed, using bundled English OCR data, and a library index speeds up repeated searches, updated after archives are added or replaced.
It runs entirely locally, with no API key or model download required for the default hashing embedder, and is installed from PyPI with Python 3.10 or newer.