acatome-store 0.3.0

Persistent storage, dedup, metadata queries, and semantic search for acatome bundles

acatome-store

Persistent storage, deduplication, metadata queries, and semantic search for scientific paper bundles.

Features

• SQLAlchemy ORM — portable across SQLite, Postgres, MySQL
• Refs + Papers split — identity table (refs) separate from ingested content (papers)
• Citation graph — directed citing → cited edges, works for ingested + stub papers
• Supplements — ingest supplementary PDFs with scoped block retrieval
• Retractions — flag papers as retracted with notes
• Vector search — ChromaDB (default) or pgvector (zero text duplication)
• CLI — acatome-store command for ingest, query, retract, and stats

Installation

uv pip install -e .

With Postgres support:

uv pip install -e ".[postgres]"

Usage

from acatome_store import Store

store = Store()
ref_id = store.ingest(bundle_path)
paper = store.get(ref_id)
results = store.search("transformer attention", top_k=5)

CLI

acatome-store ingest /path/to/bundle.acatome
acatome-store stats
acatome-store retract doi:10.1234/fake --note "Fabricated data"

Testing

uv run python -m pytest tests/ -v

License

LGPL-3.0-or-later — see LICENSE.