Quanta
Columnar OLAP, vectorized

Quanta is a vectorized, columnar analytical database built for sub-second queries over petabyte tables. Postgres wire protocol, native S3 storage, and a query planner that thinks in SIMD.

  • Postgres-compatible
  • Separated storage + compute
  • Open-source core
quanta-shell — 1.2B rows scanned
quanta> SELECT region,
     ->        approx_percentile(latency_ms, 0.99) AS p99,
     ->        count(*) AS events
     -> FROM events_2026
     -> WHERE ts > now() - INTERVAL '7 days'
     -> GROUP BY region ORDER BY p99 DESC;

 region   | p99   | events
----------+-------+-------------
 us-east  | 142.0 | 418,902,114
 eu-west  | 168.5 | 311,447,820
 ap-south | 201.3 | 269,118,557

(3 rows) — scanned 1,204,991,002 rows in 0.38s
quanta> _

Powering analytics at data-heavy teams

HelixNorthstarCadenceVergeAtlasLumina
The engine

Built for scans norow store can survive.

Quanta separates storage from compute, compresses columns 12x, and executes queries in vectorized batches — so a full-table aggregate costs cents, not an afternoon.

Vectorized execution

The planner compiles each query into SIMD-friendly batches of 65,536 values, saturating modern CPUs and cutting per-row overhead to near zero. Aggregations that crawl in a row store finish before the spinner appears.

Columnar storage

Per-column encoding and 12x compression mean you read only the bytes a query touches.

Storage on S3

Decoupled compute reads directly from object storage. Scale readers to zero between queries.

Postgres wire protocol

Connect with psql, dbt, or any Postgres driver. No new client, no rewrite.

Time-travel queries

Query any table as of a past timestamp. Snapshots are immutable and free to keep.

Measured, not marketed

1.2B
Rows scanned/sec/core
12x
Avg compression
0.38s
Median analytical query
100PB
Largest deployment
Operate with confidence

Production-gradefrom the first node.

Durability, observability, and access control are not add-ons. They ship in the engine, audited and on by default.

Exactly-once ingest

Streaming ingest from Kafka and Kinesis with idempotent writes and replay from any offset.

Query observability

Every query emits a flame graph, bytes scanned, and cost — exportable to OpenTelemetry.

Row-level security

Policy-based access control enforced inside the planner, not bolted on at the app layer.

Zero-downtime upgrades

Rolling node replacement with no read interruption. Upgrade a 100-node cluster mid-traffic.

Pricing

Pay for compute you use.

Storage is billed at object-store rates. Compute is per second, scaled to zero when idle. No per-seat tax.

Community

The open-source engine, self-hosted.

$0
  • Single-node engine
  • Full SQL + Postgres protocol
  • Columnar + vectorized core
  • Community Slack
Most popular

Cloud

Managed, separated storage and compute.

$0.08/compute-hr
  • Auto-scaling compute pools
  • S3-backed storage
  • Time-travel + snapshots
  • Streaming ingest
  • 99.95% SLA

Enterprise

For regulated, petabyte-scale fleets.

Custom
  • Dedicated VPC deployment
  • BYOC + on-prem option
  • Row-level security + SSO
  • Named solutions architect
  • SOC 2 + HIPAA

Engineer's questions, answered.

Is Quanta really Postgres-compatible?

Quanta speaks the Postgres wire protocol, so psql, dbt, Metabase, and any Postgres driver connect unchanged. We support the analytical subset of SQL plus our own approximate and time-travel extensions.

How does it stay fast at petabyte scale?

Columnar storage with 12x compression means queries read only the columns and row groups they touch. The vectorized engine then processes those batches with SIMD, so scan time grows with the data you actually query, not the table size.

Can I separate storage from compute?

Yes — that is the default in Quanta Cloud. Tables live in S3-class object storage; stateless compute pools scale up for a query and down to zero between them, so idle clusters cost nothing.

What happens to my data during an upgrade?

Upgrades are rolling. Nodes are replaced one at a time with no read interruption and no manual failover, so a 100-node cluster upgrades mid-traffic without a maintenance window.

Is the engine open source?

The single-node Community engine is open source under Apache 2.0. Cloud adds managed auto-scaling, separated storage, streaming ingest, and the SLA.

Point your hardest query at Quanta.

Spin up a managed cluster in under two minutes, or download the open-source engine and run it tonight.