Structured and semantic search
FTS5 handles exact keyword retrieval while ChromaDB adds vector search over long text columns.
One local database
for agents and apps.
No cloud services to wire together, no clusters to manage, no data leaving the machine.
One embedded stack for agent-local data.
Hybrid ranking
Keyword and vector results are fused with reciprocal rank fusion for practical agent retrieval.
Graph-aware storage
SQLite-backed graph nodes and edges support traversal, shortest path, PageRank, and communities.
Local analytics
DuckDB syncs from SQLite so agents can run analytical SQL without leaving the machine.
Self-healing journal
Writes are journaled and replayed into vector and analytics indexes with reconciliation for drift.
Local-first by design
All engines are embedded. Data stays on device with no hosted database service required.
Architecture
Four query engines behind one Python API.
HybridDB stores truth in SQLite, indexes text into FTS5, embeds long text into ChromaDB, syncs analytics tables into DuckDB, and exposes graph algorithms over SQLite-backed node and edge tables.
SQLite + FTS5
The source of truth for rows, schemas, full-text search, and write-ahead logging.
ChromaDB vectors
Persistent collections per long-text column using 384-dimensional MiniLM embeddings.
DuckDB analytics
Columnar analytical queries over local tables synchronized from the journal.
Graph tables
Nodes and edges backed by SQLite with traversal and centrality algorithms.
The foundation for anything an agent needs to remember, search, or analyze.
check_circleRAG and knowledge retrieval — MessageStore and MemoryStore use hybrid keyword + vector search to recall conversations and observations.
check_circleEmail CRM — GmailCache and EmailStore index emails with full-text and semantic search for instant retrieval.
check_circleStructured data apps — AppBuilder lets agents create and query custom data tables backed by HybridDB.
check_circleGraph context — Entity relationships, memory provenance, and dependency chains through built-in graph traversal.
check_circleLocal analytics — DuckDB layer enables analytical queries over any registered table without exporting data.
Compared to alternatives
Not another vector database. A local agent data layer.
Chroma, Qdrant, Weaviate, and Milvus are strong retrieval systems. HybridDB is for the smaller but common problem: one embedded store for the agent's own local working data.
Versus vector databases
HybridDB does not try to beat Qdrant or Milvus at distributed scale. It wins when the app needs rows, FTS, vectors, analytics, and graph context in one local process.
Versus SQLite extensions
SQLite FTS and vector extensions are powerful pieces. HybridDB packages the retrieval workflow: schema tracking, hybrid ranking, vector sync, analytics, and repair utilities.
Versus DuckDB
DuckDB is excellent for analytics. HybridDB uses DuckDB as one layer while keeping transactional app data and retrieval indexes together.
Versus graph databases
Neo4j is the right choice for deep graph workloads. HybridDB provides good-enough graph relationships for memory, provenance, dependencies, and local agent context.
Use it when
Use HybridDB when local control matters more than cloud scale.
radio_button_checkedUse it for local-first agents, personal knowledge stores, and app-builder data where your app controls table and column names.
radio_button_checkedDo not pitch it as a distributed vector database, hardened SQL sandbox, or enterprise graph platform.
radio_button_checkedAnalytics and graph features are useful embedded layers, not replacements for dedicated warehouse or graph infrastructure.
radio_button_checkedThe clearest CTA is not 'host your vectors'; it is 'give your agent one inspectable local brain for data and retrieval.'
Give your agent one local store for memory and search.
Start with SQLite-shaped data, then add semantic retrieval, analytics, and graph context without introducing external infrastructure.
Build a local agent store