# LakeSail

> LakeSail is a cloud data and AI platform for teams that want faster insights, lower costs,
> and more control. Built for both humans and AI agents, it helps you extract meaning from data,
> reason at scale, and make decisions with confidence. LakeSail runs SQL and Python workloads in
> your own AWS account (BYOC). Under the hood, Sail is a high-performance Rust-based engine with
> Spark Connect compatibility, delivering broad support for Spark SQL and PySpark with simple,
> transparent flat-rate pricing.

## What makes LakeSail different

LakeSail is a fully managed data and AI platform that runs entirely inside your AWS account (BYOC),
giving you complete control over your infrastructure, IAM policies, and encryption keys from day one.
Your data stays in your cloud, with built-in sovereignty, governance, and access control through SSO,
role-based access, and granular permissions.

LakeSail eliminates operational overhead by provisioning compute automatically for each job inside
your VPC, scaling with the workload, and releasing resources as soon as the job finishes. There are
no idle nodes to manage, and you can choose the exact AWS instance types for each workload. It
supports both standalone mode for lightweight jobs and cluster mode for larger-scale workloads, with
scale-to-zero economics so you only pay for active compute.

The platform also includes built-in job orchestration, so you can create SQL and Python jobs, run
them on a schedule or on demand, and manage dependencies in one place. Every job is versioned, with
full run history for every execution, and LakeSail works with orchestrators like Airflow and Dagster
if you want to integrate it into your existing stack.

Under the hood, LakeSail is powered by Sail, a high-performance query engine written in Rust with
Apache Arrow-native execution. By avoiding the JVM, garbage collection pauses, and Py4J serialization
overhead, Sail delivers fast, efficient execution for SQL and Python workloads. Python UDFs run
natively inside the engine via PyO3 with zero-copy data sharing, and the engine starts in seconds
rather than minutes.

LakeSail natively supports Delta Lake and Apache Iceberg, and integrates with Glue Catalog, Unity
Catalog, Iceberg REST Catalog, and OneLake. Existing Spark Connect workloads run without code changes,
making adoption straightforward for teams already using Spark-based tools.

Performance is where LakeSail stands apart. In the ClickBench benchmark, Sail running on a single
node with 16 vCPUs and 32 GiB of memory outperformed Databricks’ best configuration of 32x X-Large
nodes with 256 vCPUs and 1,952 GiB of memory, as well as Snowflake’s published result. That means
Sail achieved better performance than Databricks using 1/16 as many vCPUs. On the derived TPC-H
benchmark (SF100, Sail 0.7.0 vs Spark 4.2.0), Sail completes the workload roughly 10x faster than
Apache Spark, with a modeled cost reduction of up to 98%. Full results:
https://docs.lakesail.com/sail/latest/introduction/benchmark-results/

## Pricing

- Community: Free, Apache 2.0 open source.
- Managed: $0.01 per vCPU-hour + $0.002 per GiB-hour, plus your AWS compute cost billed directly to your account.
- Enterprise: Custom pricing, with on-premises deployment available. Contact sales.

## FAQ

Q: Why is managed Spark so expensive on other platforms?
A: Most managed Spark platforms add a proprietary usage unit on top of your cloud bill, at a rate they set and can change. That markup can range from 2 to 4x your raw compute cost, and it rarely shows up clearly in your bill. LakeSail charges transparently for compute resources directly at hardware rates, with no multiplier.

Q: Is my data stored in LakeSail's infrastructure?
A: Your data files stay in your AWS account. LakeSail provisions EKS clusters and S3 workspace buckets into your VPC; query results and Glue catalogs live in S3 buckets you own. To manage that infrastructure, LakeSail's control plane assumes a scoped IAM role in your account (limited to `lakesail-*` resources) and reads CloudWatch metrics for the cluster dashboard. Job metadata, cluster configuration, and infrastructure state are stored in LakeSail's control plane, not your data files or query results. You can revoke access at any time by deleting the CloudFormation stack. Enterprise on-premises deployments run entirely in your own infrastructure.

Q: How long does migration actually take?
A: For most teams: one config line. LakeSail is Spark Connect compatible — swap `sc://lakesail.your-account.aws` as the remote URL and your existing PySpark code runs unchanged. Zero rewrites. Zero cost. For edge cases, the team typically ships fixes in the very next Sail release. Despite the easy switch, most teams migrate incrementally at a pace their company can handle, which we agree is a wise approach.

Q: What's included in the managed platform fee?
A: Cluster provisioning, autoscaling, monitoring, the engine license, job scheduling, cost dashboard, MCP server, and Slack/email support. You pay EC2 to AWS directly; the platform fee goes to LakeSail.

Q: Do you offer a free trial?
A: Open source is always free. Managed platform customers get a 14-day free trial. No LakeSail fees. Just your hardware. After that, $0.01 per vCPU-hour and $0.002 per GiB-hour will be applied to your compute usage. Your access to the platform stays open even after the trial ends. Run your workloads, see your savings, then decide. For custom enterprise setups or migrations we offer flexible enterprise support options and white glove service. Please reach out to us at https://lakesail.com/#talk-to-us.

Q: How does the up to 98% cost reduction hold up for my workload?
A: The 98% figure is modeled from the derived TPC-H SF100 benchmark (Sail 0.7.0 vs Spark 4.2.0): Sail ran the workload 10x faster, which alone is a 90% cost reduction, and its 26 GB peak memory fits an instance 1/4 the size, multiplying out to nearly 98%. The instance-size factor assumes memory drives instance sizing and pricing. Because workloads vary, your number might be higher or lower. Benchmark your own workloads and see exactly where you land, before any commitment.

Q: Is there a cost to migrate?
A: No. LakeSail is Spark Connect compatible, so there are no code rewrites, no consultants to hire, and no professional services fees. The only cost is your time — typically a few hours to swap the endpoint and validate.

## What LakeSail is NOT

- Not a database. LakeSail is a compute layer; your data stays in your cloud storage (S3, Iceberg, Delta Lake).
- Not a SaaS hosted service. LakeSail deploys into your AWS account (BYOC); your data never leaves your VPC.
- Not a Spark fork. Sail is a Rust-native engine with Spark Connect compatibility — different codebase, same interface.
- Not Databricks or Snowflake. Different architecture, deployment model, and cost structure.

## Company

LakeSail, Inc. was founded in 2023 by Shehab Amin (Co-Founder & CEO) and Heran Lin (Co-Founder & CTO).
Leadership: Shehab Amin (Co-Founder & CEO), Heran Lin (Co-Founder & CTO), Peter Jackson (Chief Strategy
Officer), Everett Roeth (Head of Go-to-Market), Mike Price (Head of Customer Experience), Alexy Khrabrov
(Head of Ecosystems), and Dave Osborne (Operating Advisor). Full bios: https://lakesail.com/about/

## Customers and validation

- Used by teams at Microsoft, JPMorgan Chase, HPE, Société Générale, and Adyen.
- Quoted by Andrew Lamb (Apache PMC, Apache Arrow & DataFusion / InfluxData), Jordan Taylor (Data Engineer, Office for National Statistics), and Santosh Pingale (Principal Engineer, Data, Adyen | Sail Contributor).
- Backed by independent benchmarks: derived TPC-H runs (Sail 0.7.0 vs Spark 4.2.0) show Sail roughly 10x faster than Apache Spark with a modeled cost reduction of up to 98%. ClickBench results (per the performance page) show single-node Sail outperforming Databricks' 32x X-Large config.

## Product

- [Platform Overview](https://lakesail.com/product/): Full-featured data platform for engineering, analytics, and AI
- [Data Engineering](https://lakesail.com/product/data-engineering/): ETL/ELT pipelines with SQL, Python, or Spark-compatible code
- [Agentic Infrastructure](https://lakesail.com/product/agentic-infrastructure/): Governed agent compute with MCP, sandboxed execution, and elastic swarms
- [Spark Connect Compatibility](https://lakesail.com/product/spark-connect/): Drop-in replacement for Spark SQL and PySpark
- [Python Workloads](https://lakesail.com/product/python-workloads/): PySpark-compatible DataFrame API with zero-copy UDFs via PyO3

## Resources

- [Blog](https://lakesail.com/blog/): Engineering deep dives and product updates
- [Documentation](https://docs.lakesail.com): Guides, tutorials, and API references

## Optional

- [GitHub (Sail)](https://github.com/lakehq/sail): Open-source query engine repository
- [Privacy Policy](https://lakesail.com/legal/privacy-notice/): Privacy and data handling