Tech News
← Back to articles

Why We Migrated from Neon to PlanetScale

read original related products more articles

In May 2025, during the same week Neon announced their acquisition, our databases went down four times. For hours. Database spin-ups, their entire value proposition, were completely disabled. Our "serverless" databases couldn't even start.

That was the final straw in our decision to migrate to PlanetScale.

Who We Are and Why Databases Matter More

At OpenSecret, we're building something unique: a confidential computing platform powered by AWS Nitro Enclaves. Our flagship application, Maple AI, provides truly private AI chat where even we can't see user data.

Here's how it works: User data is encrypted client-side before leaving their device. It travels encrypted to our servers, where it's processed inside hardware-isolated enclaves with encrypted memory. Even with root access to our servers, you can't decrypt user data. Only the verified secure enclave can.

This architecture means our database stores encrypted blobs that are meaningless without the enclave's isolated computation environment. Even our database provider cannot read our users' data. But this makes reliability absolutely critical, we can't just "peek" at the data to debug issues. When something goes wrong, we're debugging blind unless we have excellent observability tools.

We have thousands of users generating thousands of chats daily. Lawyers discussing sensitive cases. Executives planning strategy. Developers working on proprietary code. They trust us because their data is mathematically guaranteed to be private. But that trust evaporates quickly when the service is unavailable.

The Breaking Point with Neon

Neon's scale-to-zero architecture seemed clever initially, why pay for idle databases? But for an always-on AI chat application where users expect instant responses, the cold starts and spin-up delays were problematic.

The May outages during their acquisition week were devastating. Four separate incidents where databases were down for hours. The core feature we relied on, databases spinning up on demand, completely failed.

... continue reading