TL;DR: AI applications and agents on Databricks Lakebase need a trusted transactional schema, not ad-hoc tables. Quest Lakebase Modeler is a native Databricks app that lets teams create, AI-enrich, map, and deploy governed Lakebase schemas from business intent, with human approval at every step and the model as the single source of truth.
AI is only as good as the data it stands on. As organizations move from AI to experiments to AI applications (agents that take action, apps that write records, assistants that answer from live operational data), that foundation is no longer only analytical data. Increasingly, it is Databricks Lakebase, serverless Postgres for the data lake, which brings transactional workloads onto the same platform as your analytics and AI data.
That shift puts a new question in front of every team: what does the data actually look like? An agent writing orders to a table with no declared keys, or an application reading records whose relationships only exist within a developer’s head will produce unreliable results no matter how capable the model on top. Structure is what turns AI output into an AI outcome you and your organization can trust.
That is why Quest Software and Databricks have been working together to give Lakebase teams a modeled, governed foundation from the first table. Today, we are introducing Quest Lakebase Modeler, a data modeling and schema delivery application built exclusively for Lakebase. It can be installed directly from Databricks Marketplace and is built natively on Lakebase to run inside your Databricks workspace.
Databricks Lakebase changes what a schema is
For a decade, formal data modeling on Databricks was optional. Delta tables could be rewritten, schemas evolved and schema-on-read absorbed most design mistakes. On the analytical side, a wrong table meant a re-run.
Lakebase changes that. A Lakebase schema is a contract with running application code, and changing it means a migration, a rollback plan and a maintenance window. Three things follow:
- Design decisions now have operational consequences. Referential integrity, key strategy, indexes and data types all matter. Some carry direct platform costs as synced tables need well-defined primary keys to sync incrementally.
- The people making those decisions are new to them. Lakebase is being adopted by data engineers, DBAs and application developers who know Delta and pipelines well but have not been asked to design a transactional schema before.
- Existing modeling tools live outside the platform. So, teams either skip formal design or inherit migration debt by the second sprint or move the work to a desktop tool disconnected from Unity Catalog.
Meet Quest Lakebase Modeler
Quest Lakebase Modeler closes that gap natively, in the workspace where the work is already happening. It takes a team from business intent to a governed, deployed Lakebase schema in four steps.
- Create the model. Start wherever you are: author a model by hand, describe the application in plain language and let AI draft it, reverse engineer a live Lakebase database or import Unity Catalog tables and derive an operational model.
- Enrich the model. AI proposes relationships, shared domains and descriptions across the model. Each proposal shows its confidence and its reasoning, and lands in a review workbench as a recommendation, not a change.
- Map to Unity Catalog. Source mapping binds model entities and attributes to existing Unity Catalog tables and columns or generates new Unity Catalog structures for entities that do not exist yet.
- Deploy to Lakebase. DDL is generated from the model, reviewed and approved, then applied to Lakebase. Teams can deploy to the primary database or to a copy-on-write Lakebase branch, and promote through development, staging and production.
Two principles run through every step of the process.
AI with a human gate. Generation never writes to the model on its own. A person approves, rejects or edits every suggestion and edits are kept alongside the original. Every change in a deployed schema was approved by someone accountable for it. That is what lets teams use AI to move faster without letting it write production DDL unsupervised.
The model is the source. Because DDL is derived from the model rather than written by hand, the documented design and the deployed database stay the same artifact. They cannot drift apart.
What teams use it for
Quest Lakebase Modeler is built around a handful of recurring jobs:
- Designing schema for a new AI application. Describe the application, review the drafted model, refine it and deploy, with the design captured in a governed model instead of CREATE TABLE statements scattered across notebooks.
- Serving Lakehouse data to an application. Model the operational tables the application writes and the read side served through synced tables together, including the primary keys incremental sync depends on.
- Documenting a database nobody documented. Reverse-engineer a live Lakebase schema and get proposed descriptions, domains and missing relationships, including constraints that should be declared but aren’t.
- Standardizing across teams. Shared domains give one definition of a currency amount, an email address or an identifier, so seventy tables from three teams don’t arrive with three conventions.
- Changing a production schema safely. Deploy the revised model to its own Lakebase branch, an instant copy of production, and run the application against it. If the design is wrong, discard the branch instead of unwinding a migration.
See it in action
Walk through the high-level interactive demo below, from a natural language prompt to a reviewed model and generated DDL promoted to production.
Built for Databricks, delivered on Databricks
Quest Lakebase Modeler is not a desktop tool pointed to a database. It is a native Databricks application.
- Runs in your workspace. It installs from Databricks Marketplace and runs on the Databricks Apps serverless runtime, on your own compute. There is no Quest-hosted service in the data path.
- Governed by Unity Catalog. Every action runs under the signed-in user’s Unity Catalog identity, so existing grants apply. Deployed objects are visible and governed in Unity Catalog like any other Databricks object.
- AI on Databricks by default. AI-assisted features use Databricks model serving out of the box, so schema metadata stays inside the workspace. Customers can configure another provider with their own credentials. Only metadata is used; table data is never sent for inference.
- Built on Lakebase branching. Copy-on-write branches let several teams, or several revisions of one model, be validated at once without contending for a single database.
For buyers, that means no new infrastructure line item. Compute and inference run on your own existing Databricks account.
Who it’s for
- Data engineers and application developers who need a correct Lakebase schema quickly, without first needing to become modeling specialists.
- Application owners who need a governed database behind an application on a deadline, properly connected to the lakehouse.
- Platform and data architecture leads who need models that are reviewed, consistent and documented, and assurance that AI is no writing production DDL unsupervised.
- Data modelers who want to work on Lakebase inside of Databricks rather than in a disconnected desktop tool.
Quest has spent decades helping enterprises model their data, as the company behind erwin Data Modeler. Quest Lakebase Modeler brings that expertise to a single platform, natively. Together with Databricks, our aim is simple: every Lakebase application, and every AI initiative built on it, should start on a data foundation that teams can trust.
Learn more today
- Read the press release
- Find Quest Lakebase Modeler on Databricks Marketplace
- Download the datasheet
- Visit the Quest Lakebase Modeler product overview page
- Purchase today from the Quest e-store
