arXiv:2611.07971 · v1
ArcadeDB in Python: An In-Process Multi-Model Database
摘要
Some Python workflows need transactional reads and writes over documents, traversals over a graph, and similarity search over vector embeddings. In practice these are split across separate systems (a document store, a graph engine, and a vector index), and the data is copied between them and kept consistent by application code. ArcadeDB is an open-source multi-model database engine, written in Java, that supports all three access patterns and can be embedded in a host process. We present arcadedb-embedded, a Python package that runs ArcadeDB inside the Python process via JPype, using a Java Runtime Environment bundled in the wheel. One pip install gives a database with documents, a property graph, and vector search, and needs no Java installation. Query results can be returned as Python lists, NumPy arrays, pandas DataFrames, or Arrow tables. We demonstrate a hybrid retrieval workflow that chains vector search, SQL filtering, and graph traversal over one dataset. We benchmark arcadedb-embedded against SQLite, DuckDB, LadybugDB, and Chroma, and report its costs.
作者备注：12 pages. Published in the SciPy Proceedings 2026 (Proceedings of the 25th Python in Science Conference), https://proceedings.scipy.org/articles/fwvt2525
期刊信息：Proceedings of the 25th Python in Science Conference (SciPy 2026)