PlanetScale has made Neki available in platform preview. Neki is a sharded Postgres platform: every shard is a real Postgres instance, and Neki layers a router, sidecars and a control plane on top. The stated targets are hundreds of millions of queries per second and petabytes of data per database, with zero-downtime resharding. Nick Van Wiggeren announced the launch in a blog post.

Applications connect to the Neki router, which directs queries to the Postgres shards. When a shard gets hot, Neki can split it online: it creates new target shards, catches them up via replication and moves traffic through topology changes. Resharding runs while the database stays online, without application rewrites or maintenance windows.

The feature list is broad. High availability is handled by the platform, which tracks health, promotes replicas and updates routing while applications keep a single Postgres endpoint. Schema changes run as coordinated workflows across shards with per-shard methods and tracked cutover. Connection pooling is built into the sharded layer with awareness of topology, shard health and session state. Shard groups let operators assign different tables or workloads to different hardware profiles. Each shard runs as a highly available Postgres cluster across availability zones.

Data placement is explicit through a concept Neki calls data topology, mapping data to shards, shard groups and shard indexes. Cross-shard transactions with atomic commit are announced as coming soon. Explicit workflows cover imports, resharding, schema changes, cutovers and cleanup. Online version upgrades follow the same migration model: create target shard groups, replicate, cut over, retire the old ones. Zero-downtime imports copy, replicate, verify and cut over while the source keeps serving traffic. For CDC, each shard keeps its own WAL and logical replication path, compatible with standard Postgres patterns.