REAL-TIME
ANALYTICS &
SEARCH DB
The unified engine for customer-facing analytics. Sub-second performance on petabytes of data with 100x more concurrency.
VeloDB and Apache Doris help over 10,000+ enterprises to bring real-time analytics into the AI era
Challenges with real-time analytics
in the AI era
The Multi-Database Trap
One DB for search, another for complex joins, a third for dashboards. No unified warehouse — just fragile ETL pipelines duct-taping silos together.
The Real-Time Paradox
Data freshness or query speed — pick one. Sub-second SLAs slip under heavy AI workloads. Endless tuning just to keep up.
AI Concurrency Bottleneck
Agentic workflows demand high-concurrency search across structured and unstructured data. Without it, real-time RAG becomes unsustainable.
Built to solve current and emerging real-time data challenges
Four workloads, one engine. Each powered by purpose-built technologies that deliver measurable outcomes.
Customer / Agent-Facing Analytics
Insights in milliseconds with performant search and aggregations under fast-changing data
Materialized Views
Incremental materialized views replace batch ETL, refreshing complex transformations in minutes instead of hours. Data is queryable the moment it lands.
Inverted Index + Smart Indexes
Partition pruning, bucketing, pre-aggregated tables, and inverted indexes let queries touch only the data they need — keeping P99 latency under 100ms even at 10,000+ QPS.
MCP Server for AI Agents
A built-in Model Context Protocol server lets autonomous AI agents query live operational data through a standard interface — no custom integration code required.
Vectorized MPP Engine
A Cost-Based Optimizer with runtime filtering drives the MPP execution engine. Complex multi-table JOINs that take minutes elsewhere finish in seconds.
Log, Trace & Metric Analytics
Analyze and search PB-scale log, trace, and metric data effectively
Inverted Index for Full-Text
Native inverted indexes replace Elasticsearch's architecture. Full-text keyword search across 1 billion log records returns in under 2 seconds — with 5× the write throughput.
VARIANT Data Type
Schema-on-read for JSON logs. The VARIANT type auto-extracts fields as typed sub-columns without ETL, delivering 8× analytical performance and 3× compression over raw JSON.
ZSTD 1:5–1:10 Compression
Columnar storage with ZSTD compression achieves 5–10× compression ratios. Tiered storage on object storage further reduces costs by 70%, making PB-scale retention affordable.
OpenTelemetry Native
Drop-in integration with Logstash, Beats, OpenTelemetry exporters, Grafana, and Langfuse. Your existing observability stack works out of the box.
Hybrid Search for RAG
Power GenAI with a cost-effective knowledge store, leveraging hybrid search and progressive filtering
Full-Text Search (BM25)
Built-in BM25 ranking with inverted and N-gram indexes. No external search engine needed — keyword relevance scoring runs natively inside the same engine as your analytics.
Vector ANN (HNSW)
High-performance approximate nearest neighbor search with HNSW indexes. Embed and query vectors alongside structured data in the same table, the same query, the same transaction.
Columnar JSON + VARIANT
Schemaless JSON ingestion with automatic columnar sub-column extraction. Perfect for evolving knowledge base schemas — no migrations, no ETL.
Progressive Filtering in SQL
Combine SQL predicates, full-text ranking, and vector similarity in a single execution plan. Filter by tenant, rank by text, sort by embedding distance — one query, one round-trip.
Lakehouse Analytics
Minimize ETL and scale real-time OLAP with Lakehouse architecture
Open Format Support
Query Iceberg, Hudi, Hive, and Delta Lake tables directly. Federated queries join lakehouse data with real-time tables — no data movement, no copies.
Compute-Storage Separation
Scale compute nodes independently in seconds. Keep 100% of data on cost-effective S3/BLOB storage. Spin up clusters for peak traffic, shut them down when done.
CBO for Complex JOINs
A Cost-Based Optimizer with advanced statistics drives distributed multi-table JOINs. 3–8× faster than Greenplum on TPC-H, 3× faster than Trino on TPC-DS at 1TB scale.
Fast Metadata Caching
Local SSD caching of hot data and metadata eliminates the object storage latency penalty. Delivers sub-second interactive queries on petabyte-scale lakehouse datasets.
Delivering real-time performance
at scale
POWER YOUR
NEXT GENERATION
Join high-growth companies building lightning-fast search and analytics features with VeloDB.
No credit card required