The Serverless
Index-Streaming Database

You own your data, not us. BenoStreamDB is an index-overlay engine for the lakehouse. We layer reconstructible, persistent secondary indexes—HNSW/IVF vector search, BM25 full-text, and CSR graphs—directly onto Apache Iceberg tables in object storage.

⚡

Serverless by Default

Run directly against object storage via an embedded library. Zero operational overhead, scales to zero. Prefer a network endpoint? Use our Arrow Flight SQL server or REST gateways.

🧬

Overlay Indexing

It is not a storage engine you have to migrate into. Point BenoStreamDB at an existing Apache Iceberg table it does not own, and build indexes over that data in place. No data duplication.

🕸️

Graph Analytics

Unify vector search with topological relationships. Traverse large-scale CSR graph indexes natively using SQL UDFs directly on your edge tables without a separate graph database.

🏎️

Hardware Accelerated

Out-of-the-box GPU acceleration for Vector AI workloads using NVIDIA CUDA, AMD ROCm, Apple Metal, and Intel XPU. Includes TurboQuant (TQ8) for 4x index compression.

📊

SQL & pgvector

Unified DataFusion SQL interface. Execute exact or approximate nearest neighbor vector searches using familiar pgvector syntax (<->, <=>, <#>) directly over your lakehouse.

🔗

Ecosystem Connectors

Native connectors for Apache Spark (3.5/4.0), Trino with MERGE INTO support, a complete dbt adapter, and an Arrow Flight SQL server for zero-copy streaming over gRPC.

🔌

Compatible REST APIs

Features an optional HTTP search gateway that is compatible with OpenSearch (Elasticsearch 7.10) and Qdrant Vector REST APIs. Connect your existing frontends and tools effortlessly.

Why BenoStreamDB?

The Old Way (Fragmented Data Stacks)

Data Hostage Situation

SaaS vector databases force you to ingest and duplicate your data into their proprietary storage engines.

Proprietary Formats

Many modern databases force you to migrate your data into custom formats, breaking compatibility with standard lakehouse tools.

Always-On Clusters

Traditional graph and vector databases require complex distributed server clusters running 24/7, driving up your cloud compute costs.

Siloed Graph Databases

Vector databases lack structural traversal, forcing you to maintain and sync a separate graph database like Neo4j.

The BenoStreamDB Way

You Own Your Data, Not Us

BenoStreamDB queries data exactly where it already lives (S3, GCS, Azure) using standard Apache Parquet files. No vendor lock-in.

Apache Iceberg Native

BenoStreamDB is a fully native Iceberg engine. Write your data directly through us, or point us at existing tables. We write pure Apache Iceberg V2/V3 metadata, keeping your data fully readable by Spark, Snowflake, and Athena.

Embedded Serverless

Installs as a Python or Rust library running entirely in-process. You only pay for object storage. Instant scaling to zero.

Multi-Modal Execution

Execute exact vector similarity searches, BM25 full-text, and CSR graph traversals in a single, unified DataFusion SQL query.

Build Over Your Lakehouse Today

Zero cluster overhead, zero vendor lock-in. Query Apache Iceberg tables in object storage with vector search, BM25, and graph indexes in-process.

$ pip install benostreamdb

Questions, architecture feedback, or benchmarks?
Reach out directly to dev@benolabsai.com

The Serverless Index-Streaming Database