Lakesail
Founded 2023 Β· 15 employees on LinkedIn Β· 1 known investors
LakeSail offers a Rust-based data processing engine ("Sail") compatible with the Apache Spark API via the Spark Connect protocol, letting data teams run existing PySpark, Spark SQL, Delta Lake, and Iceberg workloads without rewrites. It also provides an agent layer with an MCP server, lakehouse branching, and lineage aimed at supporting AI agents for batch, streaming, SQL, and AI workloads.
Also known as LakeSail, Inc. Β· Sail
Founders & leadership
Lakesail was founded in 2023 by Everett Roeth.
Investors Β· 1
Company profile
researched Aug 2026LakeSail, Inc. develops Sail, an open-source distributed compute engine written in Rust and built on Apache Arrow and Apache DataFusion. Sail implements the Apache Spark Connect protocol, so existing PySpark DataFrame code, Spark SQL queries, and Python UDFs run against it by changing a single configuration line to point at a new endpoint, without code rewrites or data migration. The engine reads and writes Apache Iceberg and Delta Lake tables natively, with time travel, schema evolution and partition evolution, and integrates with object storage on AWS S3, Azure, Google Cloud Storage and Cloudflare R2.
The stated architectural premise is that the JVM is the limiting factor in Spark-era systems: Sail avoids JVM startup, garbage collection and memory tuning, executes queries in a vectorized columnar fashion, runs Python UDFs in-process through PyO3 rather than across an inter-process boundary, and uses stateless workers that scale to zero between jobs. LakeSail positions one engine for batch ETL, stream processing, interactive SQL and AI/ML workloads including LLM inference, embedding generation and model scoring, along with multimodal lakehouse data such as PDFs, images and video.
Alongside the engine, LakeSail ships an agent layer: a native Model Context Protocol (MCP) server for connecting MCP-compatible AI agents to the lakehouse, dynamic Python tooling, git-like lakehouse branching for isolated and reversible agent work, audit trails and lineage. Sail runs on Kubernetes with cluster-level task scheduling and resource management, and the company has said its enterprise platform will also support unikernels via Unikraft.
Founding story
The founders framed LakeSail as a first-principles response to why distributed data processing remained heavy, slow and hard to optimize, particularly when unifying data and AI workloads. They attributed persistent inefficiencies to JVM-based execution models, memory handling and weak support for multimodal workloads, and argued that Rust's maturation into a production-grade systems language with memory safety, compile-time guarantees and deep Python interoperability made a fully Rust distributed engine viable for the first time. The stated mission is to unify batch, streaming and AI compute in a single composable multimodal framework.
Business model
B2B software. LakeSail develops Sail as an open-source Rust engine and commercializes it through an enterprise offering: enterprise support, and a managed platform that runs in the customer's own cloud account ("your cloud, your data, your control"). Caplight characterizes the customer profile as subscription SaaS and usage-based SaaS.
Compute-hour based billing for the platform, described as transparent and predictable with no opaque credits, no minimum spend and no contract lock-in, with autoscaling to zero between jobs; a separate Sail Enterprise Support offering is sold to organizations.
Traction
LakeSail reports that Sail is used in production for large pipelines across multiple verticals and cites growing open-source contributions, a Slack community and community benchmarks. Caplight lists headcount of 15 as of mid-2026, up from 10 earlier in the year, and places the company in the top 5% of tracked companies on its momentum index. Named executives and advisors beyond the founding team include a chief strategy officer and a GTM advisor to the CEO.
Latest developments
Recent work centers on distributed execution robustness and the agentic layer: Sail 0.5 redesigned the control plane with task region scheduling, unified shuffle and failure recovery; Sail 0.6 moved to Arrow end to end with zero-copy UDFs and an Arrow Flight SQL server; Sail 0.7 introduced storage-backed blocking shuffle and checkpointing so compute remains stateless and elastic. LakeSail also introduced Sessions, on-demand access to a LakeSail cluster in the customer's own cloud for both people and agents, and an Agent Skills/Sail CLI feature for one-shot execution of PySpark scripts. Caplight signals report headcount growing from 10 to 15 employees during 2026.
βΈFull profile β market position, technology, go-to-market, geography, history, risks & controversies
Market position
LakeSail is a private, venture-backed infrastructure company in the AI, data and cloud infrastructure sector, positioned as a drop-in replacement for JVM-based Apache Spark and an alternative to Spark accelerators such as Photon. Caplight lists comparable companies including bodo.ai, Daft and DualBird, and classifies LakeSail's AI layer as infrastructure. Apache DataFusion PMC member and InfluxData staff engineer Andrew Lamb is quoted on the company's site describing the platform as combining native performance with managed ease of use.
LakeSail's positioning rests on keeping the Spark API surface while replacing the runtime: full Spark Connect compatibility for Spark 3.5 and 4.x means no code rewrites or table conversions, in contrast to accelerators that layer onto the JVM. Other stated differentiators are native in-process Python execution without serialization overhead, exclusive use of open formats and projects (Iceberg, Delta Lake, Spark Connect, Arrow, DataFusion) rather than proprietary formats, deployment in the customer's own cloud account with compute-hour billing and no minimum spend, and an agent layer (MCP server, lakehouse branching, lineage) built into the engine rather than added afterward.
Technology
Rust runtime with no JVM, garbage collection or serialization tax; built on Apache Arrow's columnar in-memory format and Apache DataFusion's vectorized query engine with SIMD acceleration where available. Sail implements the Spark Connect protocol for Spark 3.5 and 4.x, uses an in-house Rust SQL parser built with parser combinators and procedural macros for Spark SQL parity, and executes Python UDFs natively in-process via PyO3 with an embedded Python interpreter in the Rust worker. It provides native Iceberg and Delta Lake read/write with the formats implemented in its own planner and runtime, object storage integrations across AWS S3, Azure, GCS and Cloudflare R2, distributed execution on Kubernetes with task scheduling and failure recovery, storage-backed shuffle and checkpointing for stateless workers, an Arrow Flight SQL server, and an MCP server with lakehouse branching and audit trails.
Go-to-market
Open-source-led: Sail is distributed as the PySail package and developed publicly on GitHub, with a Slack community, issue and pull-request contributions, and community benchmarks. Commercial motion runs through enterprise support, direct contact with the team, and a website offer to book time with a LakeSail engineer for a benchmark against a prospect's existing Spark workloads, plus an early-access signup and a savings calculator comparing current Spark or Databricks compute spend to a LakeSail run rate.
Data and AI engineering teams running existing Apache Spark, PySpark, Spark SQL, Delta Lake and Iceberg workloads, including organizations seeking to reduce Spark or Databricks compute spend, and teams building AI agents that need governed access to lakehouse data.
Geography
United States; Caplight lists the headquarters as United States and Prospeo lists San Francisco, California.
History
Sail's first public release, version 0.1.0.dev0, shipped in mid-2024, accompanied by a preview and a derived TPC-H benchmark. Over the following year the project added distributed execution on Kubernetes, an in-house Rust SQL parser built for Spark SQL parity, PySpark UDF support via an embedded Python interpreter in the Rust worker, an MCP server, Delta Lake read/write, and native integrations with AWS S3, Azure, Google Cloud Storage and Cloudflare R2. Enterprise support was introduced in 2024. Subsequent releases added Spark 4.0 support (0.3), native Apache Iceberg (0.4), a redesigned distributed control plane with failure recovery and a SQL-queryable system catalog (0.5), end-to-end Arrow with an Arrow Flight SQL server (0.6), and storage-backed shuffle and checkpointing for stateless compute (0.7). A Sessions product providing on-demand cluster access for humans and agents was described in 2026.
Risks & controversies
Performance and cost claims (10x faster, 98% lower infrastructure cost) are modeled by the company from a derived TPC-H benchmark on r8g.4xlarge instances, with the site noting the savings calculator applies roughly half that reduction as a conservative estimate and that actual results vary by workload mix and cluster utilization. Earlier company benchmarks cited approximately 4x speedup and 94% hardware cost reduction, so the figures have shifted across releases. Third-party aggregator profiles disagree on basic firmographics: founding year is given as 2024 by Caplight and 2023 by Prospeo, headcount as 15 versus 1-10, and Prospeo states the company has never raised funding while Caplight records a seed round; Prospeo's revenue and valuation figures are described on that page as estimates derived from industry averages. The company is small and competes with well-resourced incumbents in the Spark and lakehouse market.
Compiled by commissioned research from 8 cited public sources β announcements, filings, and press listed under research sources below.
Key figures
latest reportedCompany-reported or press-reported figures, each dated to when it was claimed β not independently audited.
Timeline Β· 15
launches, deals, and filingsStorage-backed blocking shuffle and checkpoint moving intermediate job state to object storage so compute stays stateless and elastic.
On-demand access to a LakeSail cluster running in the customer's own cloud, for people and agents.
Zero-copy UDFs from Spark 4, the Variant type in SQL, and an Arrow Flight SQL server for Flight SQL clients.
Sail CLI feature for one-shot execution of PySpark scripts, giving agents on-demand data and AI engineering capabilities.
Redesigned control plane for distributed execution with task region scheduling, unified shuffle, failure recovery, and a SQL-queryable system catalog.
Native Apache Iceberg support and improvements to Delta Lake integration.
Native Delta Lake read and write support plus expanded object storage integration with Azure, GCS, Cloudflare R2, and AWS S3.
Sail 0.3 supports Spark 4.0 while maintaining Spark 3.5 compatibility, with faster object store performance and revamped documentation.
Improvements to temporal data handling, SQL compatibility, Parquet integration, and the MCP server.
Model Context Protocol server enabling LLM agents and humans to run Spark analytics and natural-language queries against data.
Preview release introducing support for distributed processing on Kubernetes.
LakeSail announced a dedicated, customizable enterprise support offering for Sail.
LakeSail published the 0.1 release of Sail with the PySail package and a documentation site.
Preview release of Sail announced with derived TPC-H benchmark results showing nearly 4x speed-up and 94% hardware cost reduction versus Spark using the same PySpark code.
Dated company events from announcements, filings, and press; legal rows summarize public dockets and regulator releases.
βΈResearch sources Β· 8
primary sources listed
- Lakesaillakesail.com Β· web
8 public sources were cited for this profile; the first-party ones are listed here.
Frequently asked questions
- What does Lakesail do?
- LakeSail builds Sail, a Rust-native, Spark-compatible compute engine for batch, streaming, SQL and AI agent workloads.
- Who founded Lakesail?
- Lakesail was founded by Everett Roeth in 2023.
- Who are Lakesail's investors?
- Lakesail's investors include Mango Capital.