Customer Zero: How Retool scaled its operational layer with Lakebase and cut costs by 80%

Retool gives teams a platform to safely ship apps, agents and AI-native workflows —whether they’re built in Retool, or elsewhere. As vibe-coded apps multiplied, so did the databases behind them, and running one million always-on databases became a cost, operations and governance problem. Retool became Customer Zero for Lakebase, moving Retool’s operational layer onto serverless Postgres governed by Unity Catalog. Operating costs fell 80%, provisioning dropped to milliseconds and governance now scales with every vibe-coded tool customers run in production.

A million always-on databases turned scale into a cost, operations and governance problem

Every AI-built app needs its own governed database

Retool spent close to a decade helping companies build good software, and its mission now centers on making AI development fast for builders and safe for the business. That safety question grew urgent as headlines filled with vibe-coded apps leaking sensitive data onto the open web. Every app and agent a customer prompts into existence needs its own production-grade database to hold state and shared context, which meant Retool was standing up a new database with nearly every new app.

The Retool Database, a built-in data store inside Retool, gives every new cloud organization an isolated database for persisting and sharing data across apps, workflows and agents. Because provisioning occurs in the first minutes of the customer experience, any delay or gap lands squarely on new users the moment they start building.

The $1 million problem: cost, disruption and exposure

As adoption climbed past one million databases, three pressures compounded at once. Each Retool database instance ran as an always-on Postgres cluster that kept billing around the clock even while idle, so cost scaled directly with database count. Maintenance grew heavier with every instance, and earlier vendors sometimes gave only 48 hours’ notice before windows that forced Retool to warn customers about downtime. A million separate databases also meant a million governance islands with no shared controls, an exposure risk for enterprise customers handling regulated data.

“At hundreds of customers, that was manageable,” said Ryan Wong, VP of Engineering at Retool. “At hundreds of thousands, it was untenable.”

Becoming Customer Zero for Lakebase reset how Retool provisions, scales and governs its database layer

Retool tested Lakebase on itself before shipping it to customers

Retool follows a simple rule: nothing reaches customers until the team has signed off on it as customer-first, so every capability is tested within Retool before it ships. That discipline is how Retool became Customer Zero for Lakebase, running it in production well ahead of the wider market. Moving Retool’s database layer from Supabase to Lakebase was a fundamental rethink of how Retool provisions, scales and governs databases at the product level.

“We don’t put anything in front of our customers unless we, as customers, sign off first,” Ryan said.

Serverless Postgres with scale-to-zero and API-driven provisioning

Where each Retool database instance once ran as a persistent cluster that never turned off, Lakebase changed the model entirely. “Lakebase gave us true serverless Postgres — databases that spin up in under a second and scale to zero automatically,” Ryan said. Provisioning now runs entirely through APIs, so Retool creates a new database in the same code path that creates a new Retool app, and a new organization gets an isolated, ready-to-use database the moment it starts building.

Unity Catalog brought shared governance to every database

Because Lakebase sits inside the Databricks Data + AI Platform, every database inherits Unity Catalog governance, lineage and access controls with no extra setup. The same model extends through Retool’s native Databricks connectors, letting enterprise customers build and ship apps, agents and workflows directly on their own Databricks data with role-based access enforced at the data layer rather than the application layer. As more teams build with AI, security and data access controls scale alongside them.

Lower costs, shared governance and compounding product benefits

Operating costs dropped 80% as idle databases stopped billing

Scale-to-zero rewrote Retool’s unit economics, moving the company from paying for a million always-on databases to paying only for the compute and storage actually used. That shift lowered annual operating costs by 80%, and the savings compound as Retool grows rather than scaling against it. The engineers who once kept the fleet running were freed for product work, and today, zero engineers work full-time on database upkeep, while provisioning has dropped from five to 10 minutes down to milliseconds.

“Scale-to-zero is the headline, but the real story is unit economics,” Ryan said. “It let our engineers focus on building Retool instead of reinventing infrastructure.”

Unity Catalog gave every app one governance layer

Governance became automatic from the moment anything lands in Retool, so apps inherit the organization’s existing permissions, audit trails and resource-level policies. Retool’s native Databricks connectors run under Unity Catalog, giving enterprise customers a single governance layer across data at rest in the Lakehouse, data in motion through an agent and data powering a live operational app. That shared model is what makes safe building possible at the scale Retool operates.

Every new Databricks capability becomes a Retool capability

Running on Databricks rather than beside it means Retool can turn new platform capabilities into product features as customer demand appears, and three connectors carry that forward. The Lakehouse Connector lets customers build analytics and AI apps on top of their Databricks data, which Reckitt used to speed up how fast insights reach its teams.

“We are now able to deliver consumer insights to our innovations team 50–60% faster,” said Sameen Gul, Global Product Director, Digital & Gen AI at Reckitt.

The Lakebase Connector powers operational apps that read and write transactional data in real time, from inventory management to agent-driven case routing.

“The operational layer around AI is becoming just as important as the model layer. With the Retool Lakebase Connector, our teams can respond to fleet and operations data in real time and scale without re-architecting our tools,” said Agata Rother, Operations Systems and Insights Team Lead at Wayve.

Agent Bricks gives Retool agents native access to Databricks AI tools, and a new agent-to-agent connector links Databricks Genie Agents with Retool, so insights can pass from one agent to another and trigger a human workflow without anyone moving data by hand.

“We can now move from explanation to action without losing momentum. Agent-to-agent communication between Databricks Genie Agents and Retool means the right people are responding to the right signals sooner,” said Anthony Perez, VP of Innovation and Growth Programs at MiQ.

The through-line for Retool is a build-versus-buy call that favors the customer: invest where the company can truly differentiate and buy where a partner delivers a better experience. Lakebase let Retool do exactly that, at a scale that would otherwise have demanded a much larger infrastructure team.

“Good infrastructure is invisible to our customers,” Ryan said. “They log in, build apps and deploy agents, and they never think about the database layer.”