Core concepts
- SQLite is the source of truth.
- TEXT columns get keyword search.
- LONGTEXT columns get keyword and vector search.
- DuckDB analytics and graph helpers are optional layers.
- Derived indexes can be rebuilt from local truth.
HybridDB combines SQLite rows, FTS5 keyword search, ChromaDB vectors, optional DuckDB analytics, graph context, and repairable indexes in one embedded Python package.
pip install hybriddb from hybriddb import HybridDB, HYBRID, LONGTEXT, TEXT
db = HybridDB("./my_data")
db.create_table("docs", {"title": TEXT, "body": LONGTEXT})
db.insert("docs", {
"title": "Getting Started",
"body": "HybridDB stores local rows and searchable long text.",
})
results = db.search("docs", "body", "local searchable data", mode=HYBRID) create_table, insert, querysearch, search_columnsexport_sql, import_sqlbackup, restorecheck_integrity, stats, reindex