GitHub Join the waitlist

Warehouse UX · open engine · your AWS

A Snowflake-grade warehouse. On stock Spark. In your account.

Warehouse UX on stock Apache Spark. Metered by the second. In your own AWS account.

Join the waitlist Run it locally in 2 minutes
The Flashpoint DB worksheet: SQL editor beside a results grid, with a running warehouse shown in the header.
worksheet: SQL on the left, results on the right, compute metered underneath
~30s
cold start in our tests
in-VPC
compute, state and results stay in your account
per-second
metering, no minimum, $0 compute when suspended
MIT
read the code, own the deployment
01Who this is for

Three teams, one product.

Flashpoint is for people who already run Spark, or already pay for the shape Snowflake taught everyone to expect. Step through them.

TODAY Your Spark cluster, jobs, notebooks job logs WORKSHEET QUERY PROFILE COST PER QUERY You have the engine. You are missing the warehouse.

You already run Spark

You have a cluster, jobs and notebooks. What you do not have is a worksheet, a query profile per query, or a way to see what each query cost.

A warehouse face on the Spark you already run.

Read the full breakdown →

TODAY Managed platform Your own AWS account what you actually use Same shape. A different owner and a different bill.

The bill is the problem

You pay a managed platform mostly to get the SQL surface and the profiles. The compute is the part you could run yourself, in your account, on spot capacity.

Same shape, metered on your AWS bill.

Read the full breakdown →

YOUR AWS ACCOUNT · VPC Spark your IAM, your S3 Results Parquet, your bucket analyst no third party no data export

Data cannot leave the account

Your security team treats a SaaS data platform as a third-party risk. A control plane in your own account removes the conversation entirely.

Runs in your VPC, on your IAM and your S3.

Read the full breakdown →

02What changes

You already run Spark. This gives it a warehouse face.

The engine stays Apache Spark. What changes is the layer your team works in.

Spark today

  • Notebooks stapled to a long-lived cluster
  • Job logs instead of a query profile
  • A cluster bill you cannot attribute per query

Spark with Flashpoint

  • A worksheet, and warehouses you create, suspend and resize
  • An operator-tree query profile for every query
  • Per-second metering, and a cost center beside the AWS bill
03Your stack

It plugs into what you already run.

No new engine to learn and no rewrite. A warehouse exposes one Spark Connect endpoint and one REST control plane, and everything else in your stack already speaks those.

Clients

  • Any Spark Connect client, including PySpark and notebooks
  • SQL through the REST gateway, or straight to the endpoint
  • Existing DataFrame and Spark SQL code, unchanged

Automation

  • Airflow, CI or a script drives lifecycle and queries over the REST API
  • Every query profile is a stable URL you can paste into a PR or a ticket
  • Metering per warehouse, readable next to the AWS bill
04What it costs

An hour of the smallest size.

Per compute-second, and only while a warehouse runs. A suspended warehouse bills nothing.

Snowflake $2.00 / hour 1 credit/hour at list, 60-second minimum each start
Flashpoint DB $0.08 / hour one executor on the rate card, no minimum; the driver meters separately
05Your VPC

Data never leaves your account.

Gateway, driver, executors and results all run inside your VPC, on your IAM and your S3, reached over an endpoint you open.

Data stays in your VPC YOUR AWS ACCOUNT · VPC notebook SQL client gRPC Gateway stateless REST ECS Fargate Driver · on-demand Executors · spot S3 · results
06The proof

Where the query spent its time.

Not a job log. The operators ranked by duration, the slow one first.

HashAggregate 4.8 s
BroadcastHashJoin 3.0 s
Scan parquet · orders 318 ms
Scan parquet · customers 301 ms
Exchange 155 ms
AdaptiveSparkPlan 29 ms
  • Find the slow step, not the slow job
  • Share any query as a link
  • See what the query cost
07The bill

You pay for the seconds it runs.

Metered per compute-second while a warehouse is up. Suspended warehouses bill nothing.

You pay only while it runs $0.08/h $0 running suspended · $0 running time
  • Per compute-second, no minimum
  • Suspended is $0 compute
  • Live meters beside the AWS bill
08Lifecycle

Create it, run it, suspend it.

A warehouse starts on demand and stops when idle. You never manage a node.

Warehouse lifecycle Create tens of seconds on AWS Run metered per compute-second Suspend $0 compute while idle resume on activity
09Questions

Straight answers.

How close is it to Snowflake?

It copies the shape of the product: the warehouse, the worksheet, the query profile, the per-second billing. Your data stays in open formats, so there is nothing to migrate back out of.

What does it cost?

Per compute-second on a running warehouse. Suspended warehouses bill nothing. Sizes run on a published rate card from $0.08/h for one executor. There are no credits and no per-seat fees.

Where does my data live?

In your AWS account: Iceberg tables and Parquet results on your S3, catalog in your Glue, state in your DynamoDB. Nothing leaves the account.

Can I use it from a notebook or pandas?

Yes, anything that speaks Spark Connect can attach to a warehouse's gRPC endpoint. The REST gateway is there for SQL and lifecycle calls.

Waitlist

Be there when signup opens.

Early access opens in batches. Leave an address and you will get one email when a batch opens, nothing else, and no newsletter.

One email when signup opens. Addresses are stored with the mailing-list provider and never sold or shared.