CapyDB/ docs

Blog

Notes from theno-BSdatabase desk.

Short posts on managed Postgres, disposable databases, and why smaller product surfaces are usually easier to ship.

August 31, 2026

The slow kind of broken

Postgres rarely falls over. It leans - one open transaction, one overflowed cache, one WAL segment that never shipped - and nothing tells you. This release is about the quiet failures - new defaults, three new advisories, a real guard on destructive SQL, and an index advisor that finally says what an index would buy.

August 17, 2026

The p95 of not having a database API

A viral tweet celebrated going from 480 ms to 80 ms by dropping an HTTP data gateway for direct Postgres connections. We measured what the same path costs on CapyDB - and found a bug in our own benchmark while doing it.

July 8, 2026

A database is a service unit

CapyDB is built around database cells: isolated Postgres runtimes that give each project its own storage, WAL, credentials, socket boundary, pooler, extensions, and CPU/RAM envelope.

June 10, 2026

PostGIS, threshold alerts, an MCP server, and a Terraform provider

The extension allowlist nearly doubled and extensions became manageable per database. Usage alerts now warn you before the wall. And two new official clients - MCP for agents, Terraform for infrastructure as code.

June 10, 2026

Webhooks, integrations, and a real API surface

The product grew the connective tissue everything else plugs into - signed webhooks, Vercel and Netlify integrations, Clerk user sync, project-scoped keys, and an import preflight that says no before anything breaks.

April 1, 2026

Disposable databases are not a toy

Throwaway databases make testing, review flows, and bug reproduction faster without polluting the database you actually care about.

April 1, 2026

We are very intentionally not a BaaS

If you want auth, storage, functions, and platform kitchen-sink energy, there are good tools for that. CapyDB is not trying to be one of them.

April 1, 2026

A small product surface is a feature

Every layer between you and Postgres is another thing to learn, debug, or rip out later.