Skip to main content
ToraDB organizes data as a local database directory containing tables. Each table holds documents (text and optional vectors/metadata), stored in Parquet segments with optional index sidecars for fast retrieval.

​
Flow

​
Core objects

ObjectRole
DatabaseOpen a path on disk (toradb.local(path)).
TableNamed collection of documents; created with mode="text" or mode="hybrid".
SearchResultsRanked hits from table.search or retrieval SELECT.
SearchResults.provenanceStructured JSON trace of candidate flow across retrieval tiers (when explain=True).
Analytics resultTabular output from GROUP BY / aggregates via db.sql.

​
Two ways to query

  1. Python SDK — table.search(...) for retrieval; db.sql(...) for SQL.
  2. CLI — toradb query, toradb sql, toradb tables, toradb reindex.
Indexes are built automatically on ingest where possible; use reindex and compact to rebuild BM25, HNSW, or DiskANN after bulk changes.