Enterprise Data Access
A prototype no longer has to
stay a prototype.
Apps built on Replit used to stall out as neat demos, because there was no sanctioned path into the production data that IT actually governs. A native integration with Databricks' Postgres database, Lakebase, is starting to close that gap.
The Gap
The wall between
prototype and production
Agent-driven builders like Replit can turn an idea into a working app in minutes. But at most companies, the app could only touch fake seed data — real customer records or inventory tables were reachable only through systems that had already cleared an IT review. The result: most things built on Replit never got past being a convincing demo.
That's the setup Databricks announced on its official blog is changing with a native integration into Replit. Apps built by Replit Agent can now reach real, Databricks-governed data directly, without stepping outside its security and governance framework. The database backing that production data also now ships with daily automatic backups by default, so a working prototype no longer has to remain a fragile experiment.
Under the Hood
Built on a database company
acquired 18 months ago
The integration runs on Lakebase, a Postgres-compatible database whose core technology traces directly to a 2025 acquisition.
Lakebase's engine comes from Neon, a serverless-Postgres startup. According to reporting from SiliconANGLE, Databricks agreed to acquire Neon for roughly $1 billion in May 2025, and shipped it as Lakebase within a month. Neon itself was founded in 2021 by CEO Nikita Shamgunov along with engineers Heikki Linnakangas and Stas Kelvich, known for a developer pitch built around "Postgres you can branch."
Lakebase reached general availability on AWS in February 2026, and Replit's native support now connects it to the same governed rails as Databricks' existing analytics lakehouse — putting a mutable application database and a company's analytics data under one governance boundary. Pricing isn't flat, either: usage is billed per Capacity Unit (CU)-hour, and a new Always-On tier for apps that never sleep cuts baseline capacity pricing by 25%.
How It Works
The database appears
the moment you deploy
The developer only has to think about code. Replit Agent handles the infrastructure underneath.
Let the agent build the app
Replit Agent assembles the UI and logic from a plain-language prompt. At this stage the developer doesn't need to think about which database backs it.
Lakebase spins up on deploy
The moment you publish to production, Replit Agent auto-provisions a Lakebase instance. No manual step for issuing connection strings or creating schemas.
Operations continue under governance
From there, data keeps updating under Databricks' access controls, with automatic daily backups running in the background — lowering the "one crash and it's gone" risk of running this in production.
In Practice
Who it helps, and how
The upside splits cleanly along role lines.
Engineers
Less time burned wiring up DB connections and IAM policy by hand. But since real production data is now in play, agent-generated queries still deserve a human review pass.
Business & IT leaders
Internal tools can now be built in-house through a channel IT already sanctions, so the answer to shadow apps shifts from "shut them down" to "govern them." Designing that approval process becomes the new job.
PMs & solo builders
Small internal workflow tools can be built and put into real production use without waiting on an engineering team. Solo experimental use sees almost no change.
The gap between "you can build it" and "you can keep running it" is starting to close, one automated database at a time.
What's Next
What to do next, and what to watch for
In the short term, expect internal tool-building to accelerate. The easiest place to start is read-heavy use cases — dashboards, intake forms — where write access stays limited. Next, put a review step in place for the connection and query code Replit Agent generates, before scaling it up. Finally, run the numbers on Always-On pricing early, so an always-on app's monthly bill doesn't quietly drift past what an individual plan was ever meant to cover.
Still, this isn't unqualified good news. Schemas and permissions an agent generates automatically tend to be coarser-grained than what a human would design deliberately. Being connected and being safely connected are two different problems, and how fast the integration moves says nothing yet about how auditable it will be in practice. Lakebase's Azure version also remains in beta, so companies running multi-cloud setups should expect an uneven rollout.