Turn open tables into sub-second answers

Kubernetes-native lakehouse SQL and AI query engine: 10x faster, up to 50% lower cost, 1,000+ QPS at p95 under 2s, scaling in 1-vCPU increments. No copy into a warehouse, no migration.

0 5 10 15 20 25 30 35 40 45 Queries per second Seconds Atomic vs Step-Jump Scaling Production Load Pattern Step-jump scaling with legacy engines Atomic scaling with e6data $25 $50 $75 $100 $15 $74 Wasted compute Compute saved
WHAT YOU GET IN PRODUCTION

What to expect on your workloads

up to 10x
faster queries at p95
up to 50%
lower compute cost
1,000+
QPS at p95 under 2 seconds
~99%
less egress, compute runs at the data
zero
migration and data movement
1 vCPU
smallest sizing and scaling increment
WHAT IT DOES

A lakehouse compute engine, not another warehouse

e6 Query engine runs SQL and AI workloads against your existing storage: S3, ADLS Gen2, or Iceberg-managed data. Nothing is copied into a separate system first.

It is built for the three things that make modern analytics hard. High concurrency, when many people run many queries at once. Complex SQL, with joins, window functions, and aggregations over large fact tables. Cost discipline, so you pay for what you use instead of idle infrastructure.

Teams point it at the tables they already govern and keep every dashboard, notebook, and agent exactly where it is.

QUERY PATHno data movement
your SQLe6 engineyour tables
CONCURRENCY1,000+ QPS
LATENCYp95 < 2s
SCALING UNIT1 vCPU
COORDINATORnone
DATA MOVEDzero
WHERE IT FITS

One engine between your tools and your tables

Dashboards, notebooks, and agents keep their interfaces. The engine executes their heaviest work where the data already lives, in your cloud or your VPC.

YOUR TOOLSunchanged
BI dashboardsnotebooksAI agentsDatabricksSnowflake
e6data QUERY ENGINE
SQL executionvector searchautoscalingguardrails
Kubernetes-native · decoupled services · scales by 1 vCPU · no coordinator single point of failure
YOUR OPEN TABLESany lakehouse · any catalog · any cloud, public or private
IcebergDeltaHudiS3ADLS Gen2
Two ways to run it: serverless, where we manage the compute, or in your VPC, inside your own Kubernetes cluster. Your data stays in your cloud either way.
HOW IT WORKS

Six things the engine does differently

Each one is a design decision you can check against your own workload during a trial.

01 Sizing and scaling by the vCPU +

Capacity moves in 1-vCPU increments instead of doubling a cluster. A 20% rise in load costs about 20% more, so you stop paying for headroom you provisioned and never used.

proof: up to 50% lower compute cost
02 No coordinator to bottleneck +

Decentralized services communicate through defined contracts, so there is no central driver to queue behind and no single point of failure. Concurrency scales out rather than backing up.

proof: 1,000+ QPS with p95 under 2 seconds
clientexecutors × ntables
03 Complex SQL at full grain +

Joins, window functions, and aggregations over large fact tables run without down-sampling or pre-aggregating first. The queries your analysts already wrote keep working.

proof: zero query rewrites
JOINSfull support
WINDOW FUNCTIONSfull support
AGGREGATIONSlarge fact tables
REWRITES REQUIREDnone
04 SQL and AI on one engine +

Vector search runs on the same tables as your SQL, with cosine similarity for semantic lookups. There is no second store to sync, govern, and pay for.

proof: no separate vector database
SQL+ vectorone engine
05 Autoscaling with a floor and a ceiling +

Set the bounds and executors track query load between them. No manual tuning between peak and quiet hours, and no latency spike while capacity catches up.

proof: scales in 1-vCPU increments
QUIET~15 vCPU
NORMAL~60 vCPU
PEAK~144 vCPU
06 Guardrails on runaway queries +

Set thresholds per cluster and log, alert, or cancel a query in real time before it burns through compute. Cost control happens during the query, not in next month’s invoice.

proof: real-time cancellation
01threshold set per cluster
02query exceeds threshold
03log · alert · cancel
WORKLOADS

Built for your heaviest workloads

Move one workload at a time after proving the value for yourself.

01

Dashboards

BI dashboards that stay fast when everyone opens them at once, served straight off your open tables.

proof: 1,000+ QPS with p95 under 2s
02

Ad-hoc and interactive

Exploratory queries on full-grain data, without rationing which questions are worth asking.

proof: up to 10x faster queries (p95)
03

Scheduled and batch

Reports and jobs that run on compute sized to the query, so idle capacity stops billing you.

proof: up to 50% lower compute cost
04

Embedded analytics

Customer-facing analytics inside your product, holding SLAs at production scale.

proof: 15M customer-facing queries a day at p95 1.2s
WHY NOT JUST USE WHAT WE HAVE?

Add it where the bill hurts, keep everything else

e6data runs alongside Databricks, Snowflake, and Trino on the same tables. You move workloads one at a time, and the exit is re-pointing rather than re-platforming.

E6DATA QUERY
WAREHOUSE COMPUTE
OPEN-SOURCE ENGINES
Scaling granularity
1 vCPU
T-shirt sizes, step jumps
node or pod sized
Coordinator
none
central
driver or coordinator
Where data lives
your open tables
loaded into the warehouse
your open tables
Migration to adopt
none
ingest and re-model
none, but you operate it
Who operates it
serverless or your VPC
vendor
your platform team
Concurrency proof
1,000+ QPS, p95 < 2s
varies by warehouse size
varies by cluster size
PROOF

15M customer-facing queries a day at 60-75% lower TCO

A Nasdaq-listed SaaS leader serves production dashboards straight off open tables, running side by side with the platforms already in place. The bill moved and the SLAs did not.

MEASURED IN PRODUCTION
QUERIES PER DAY15M
LATENCYp95 1.2s
CONCURRENCY1,000+ QPS
TCO CHANGE60-75% lower
See the TPC-DS benchmarks →

“We achieved 1,000 QPS concurrencies with p95 SLAs of < 2s on near real-time data & complex queries. Other industry leaders couldn’t meet this even at a far higher TCO.”

CHIEF OPERATING OFFICER · CHARGEBEE
PRICING

You pay for vCPU consumption, not unused compute

Compute is billed on vCPU consumption, sized in 1-vCPU increments. There is no per-query surcharge and no charge for idle headroom, because capacity tracks load instead of stepping up in blocks.

Book a demo for pricing.

INTEGRATIONS

It plugs into the stack you already run

TABLE FORMATS
IcebergDeltaHudi
STORAGE & CLOUD
S3ADLS Gen2GCSAWSAzureGCPon-prem
CATALOGS
any catalogexisting governance
CLIENTS
JDBCODBCPythonBI tools
RUNS ALONGSIDE
DatabricksSnowflakeTrinoSageMakerMicrosoft Fabric

Query engine questions

Plain answers for the evaluator in the room.

Do I have to migrate anything?+

No. The engine reads your existing tables in place, so there is zero data movement and no query rewrites. Your catalogs, governance, and BI connections stay as they are.

How does it hold up at high concurrency?+

1,000+ QPS with p95 under 2 seconds, in production. There is no coordinator acting as a single point of failure, so parallel queries fan out instead of queueing.

What does it do to my bill?+

Capacity is sized in 1-vCPU increments, so a 20% rise in load costs roughly 20% more rather than doubling a cluster. Customers see up to 50% lower compute cost on the workloads they move.

Can it handle complex SQL, not just simple scans?+

Yes. Joins, window functions, and aggregations over large fact tables are the design target, at full grain and with no pre-aggregation step.

Where does it run?+

Serverless, where we manage the compute, or inside your own Kubernetes cluster in your VPC. Both keep your data in your cloud, and on-prem, hybrid, air-gapped, and sovereign environments are supported.

What happens to a runaway query?+

You set thresholds per cluster, and the engine can log it, alert, or cancel it in real time before it consumes more compute.

Book a demo on your own workloads

Reach out to book a demo, share challenges you're facing, and tell us how this fits into what you're currently working on or thinking about.

Prefer to self-serve? Problems we're solving

An actual person replies. By submitting, you acknowledge your personal information will be processed in accordance with our Privacy Policy.