The unified real-time database
for observability

Start Free Trial →Meet an engineer

LIVEMySQL wire

Why did checkout latency spike?

WITH spans AS (

SELECT

service_name,

operation_name,

trace_id,

duration_ms,

error

FROM trace_spans

WHERE timestamp >= NOW() - INTERVAL 15 MINUTE

)

SELECT

service_name,

COUNT(*) AS trace_count,

PERCENTILE(duration_ms, 0.95) AS p95_ms,

100.0 * SUM(error) / NULLIF(COUNT(*), 0) AS error_rate_pct

FROM spans

GROUP BY service_name

ORDER BY p95_ms DESC;

p95 Latency (all services)

842ms

↑ 78% vs prior 15 min

Error Rate (all services)

3.2%

↑ 3.2 pp vs prior 15 min

P95 Latency by Servicems

checkout-api

842

payment-svc

612

inventory-svc

298

user-svc

184

shipping-svc

162

02505007501000

Query Mix (by count)

SELECT61%
INSERT18%
UPDATE12%
DELETE9%

Top Slow Trace · 842 mstrace_id 9f3a…c7b2

checkout-api

GET /checkout

214 ms

→

payment-svc

Charge

312 ms

→

inventory-svc

Reserve

186 ms

→

db: orders

SELECT orders

130 ms

Products

The Chunking and Embedding Cookbook for Production Context Engineering

Use Apache Doris for production-ready context retrieval: chunking strategy, embedding selection, and vector index choices.

Learn more

Customer Stories

From AI labs to global scale

AI Observability

PB-scale AI logging at 70% lower cost

Replaced a Grafana Loki stack with an Apache Doris–based logging system to store and query model training and inference logs at petabyte scale.

One unified lakehouse, 6× faster queries

Consolidated Presto, Druid, Kudu, and Trino into a single Apache Doris + Paimon lakehouse running across 40+ clusters on petabytes of data.

Real-time DeFi analytics across 190+ chains

Migrated from PostgreSQL to VeloDB Cloud to serve 10M+ users with real-time on-chain analytics spanning 190+ blockchains and 500+ DEXs.

Billion-scale hybrid search, 7× faster

Built hybrid search on Apache Doris 4.0 — combining filters, BM25, and vector similarity over 1B+ embeddings — on a single server.

10,000+ Enterprises Trust Apache Doris and VeloDB

Why VeloDB

Built for the workloads that
break ordinary databases

01 · Real-Time Performance

Real-Time performance under real-world conditions

Sub-second latency under concurrency, heavy updates, and complex joins — all at once.

Most databases benchmark on clean, single-table scans. Real workloads often face the following challenges: thousands of concurrent users, data that changes by the hour, and multi-table joins. VeloDB is engineered to hold sub-second latency through all these challenges, so dashboards, data products, and AI apps stay fast as the workloads scale.

Diagram: concurrency, freshness, and complexity pressures converging into the VeloDB engine while p99 latency stays flat and under one second as load rises.

02 · Unified

Unified analytics, search, and AI retrieval

One engine for multimodal data, queried in standard SQL.

Structured tables, semi-structured JSON, full-text, and vector embeddings are stored in one place and queried with a single engine and a single SQL statement. Reduce the number of databases you have to manage and maintain so you can focus on innovation.

Diagram: one SQL statement combining structured filters, JSON path access, full-text BM25, and vector search in a single VeloDB query — replacing a search cluster, a vector database, and the ETL between them.

03 · Cost

Cost-efficient by design

Efficiency from the architecture and up.

Most analytics costs come from running and maintaining many separate systems. VeloDB lowers those costs by simplifying the stack. It replaces the search database, vector store, Redis cache, and surrounding ETL layers with built-in caching and materialized views. VeloDB Cloud separates storage and compute while integrating with open lakehouse, so compute scales elastically, and bursty agent workloads only cost you while they run. On top of that, Doris uses columnar ZSTD compression and tiered object storage to bring down storage costs.

Before-and-after diagram: five separate systems (OLAP, search cluster, vector DB, Redis, ETL) plus idle compute collapsing into one VeloDB engine with elastic compute that scales to zero — about 60% lower total cost of ownership.

04 · Open

Open by default

Truly open, fully compatible with the open-source Apache Doris.

VeloDB is built on Apache Doris and stays fully compatible with it. You get the same SQL, the same MySQL wire protocol, and the same connectors as the open-source project, so anything you build on Apache Doris runs on VeloDB without changes and ports back just as easily. VeloDB also integrates with open catalogs like Polaris and Unity, and open table formats like Iceberg, Hudi, and Delta.

Diagram: VeloDB and Apache Doris joined by a two-way arrow (same SQL, same MySQL wire protocol, same connectors), with spokes to open table formats (Iceberg, Hudi, Delta, Paimon) and open catalogs (Polaris, Unity).

Solutions

One engine, four workloads,
zero ETL between them

ANALYTICS

Real-Time Analytics

Sub-second dashboards and data products on petabytes of data at any concurrency.

Materialized ViewsInverted Index + Smart IndexesMCP Server for AI AgentsVectorized MPP Engine

Data Warehousing

Sub-second analytics on open lakehouse formats with no vendor lock-in.

Open Format SupportCompute-Storage SeparationCBO for Complex JoinsFast Metadata Caching

Observability in the AI Era

The most cost-effective alternative to Elasticsearch observability.

Inverted Index for Full-TextVariant Data TypeZSTD 1:5–1:10 CompressionOpenTelemetry Native

Context Engineering

Hybrid search and fresh context for RAG, agents, and LLMs.

Full-Text Search (BM25)Vector ANN (HNSW)Columnar JSON + VariantProgressive Filtering in SQL

The unified engine at the center of your data ecosystem

Unify data ingestion, lakehouse access, analytics, search, and AI workloads in one engine.

Power BI

Tableau

Looker

Superset

Transform

dbt

Flink

Observability

OpenTelemetry

Grafana

Langfuse

AI / ML

RAY

LangChain

LlamaIndex

Dify

The Unified Analytical Engine

INTERFACE

SQL

VECTORS

AGGREGATE

UNIFIED DATA STACK

Columnar

Vectorized

High Concurrency

Real-time

CORE

MPP Execution

CBO

Materialized Views

DATA SOURCES

Databases

MySQL

PostgreSQL

MongoDB

Streams

kafka

Confluent

Data Lake

Amazon S3

Google Cloud Storage

Azure Data Lake

SaaS Connectors

Fivetran

Airbyte

Hadoop HDFS

DATA LAKEHOUSE

Catalogs

Snowflake Open Catalog

Databricks Unity Catalog

AWS Glue

Amazon S3 Tables

Iceberg REST Catalog

Hive Metastore

Open Table Formats

Iceberg

Delta Lake

Apache Hive

LanceDB

Deploy Anywhere

AWS

Google Cloud

Microsoft Azure

Kubernetes

On-prem

Ship on the unified real-time database

Analytics, search, and AI retrieval in one engine — open source, MySQL-compatible, runs in your cloud.

Start free Meet an engineer

Recent News

This website uses cookies to enhance your browsing experience by collecting information about your interactions. By clicking Accept, you are agreeing to our use of cookies as described in our Cookie Policy.

The unified real-time databasefor observability

From AI labs to global scale

PB-scale AI logging at 70% lower cost

One unified lakehouse, 6× faster queries

Real-time DeFi analytics across 190+ chains

Billion-scale hybrid search, 7× faster

Built for the workloads thatbreak ordinary databases

Real-Time performance under real-world conditions

Unified analytics, search, and AI retrieval

Cost-efficient by design

Open by default

One engine, four workloads,zero ETL between them

Real-Time Analytics

Data Warehousing

Observability in the AI Era

Context Engineering

The unified engine at the center of your data ecosystem

Ship on the unified real-time database

Recent News

The unified real-time database
for observability

Built for the workloads that
break ordinary databases

One engine, four workloads,
zero ETL between them