// the find
neondatabase/neon
Neon: Serverless Postgres. We separated storage and compute to offer autoscaling, code-like database branching, and scale to zero.
Neon is a PostgreSQL fork that separates storage and compute, replacing the standard storage layer with a distributed pageserver and WAL safekeeper cluster. This lets compute nodes be stateless and disposable — spin up, scale to zero, branch instantly. It's for teams that want real Postgres semantics with the operational model of a cloud-native database, and for developers who want Git-like database branching for migrations and testing.
The branching model is genuinely useful: copy-on-write database branches at any LSN means you can test a migration against a real production-sized dataset without duplicating storage. The storage/compute split is architecturally clean — stateless computes mean autoscaling is a matter of adding nodes, not resharding. The WAL safekeeper quorum is a proper durability story rather than a bolt-on. Codebase is well-structured Rust with clear module boundaries between pageserver, safekeeper, and control plane.
Running this yourself is non-trivial: you're building a patched Postgres fork plus several Rust services, and the local dev setup is Linux-first with macOS being a second-class citizen. The scale-to-zero cold start latency is real — spinning up a compute node is not instant, which matters if you're using this for latency-sensitive prod workloads. Logical replication support has rough edges because the WAL is managed outside stock Postgres, so extensions that hook deep into the storage layer can break. Self-hosted operation has almost no documentation beyond the dev quickstart; if you're not using Neon's managed service, you're largely reading code.