ClickHouse Acquires Xata: Comments news update

The original has a clear section order, so I'll follow that structure and check the funding template only for constraints that change the prose.The branch list in the source paste is run together, so I’m checking the original page for those names and counts before writing them into the article.The “badge 13” string is an icon title, not a branch count. I’ll use the visible branch names, the “3 active” label, and the other figures from the post.Xata is joining ClickHouse to bring agentic capabilities to ClickHouse Managed Postgres, Monica Sarbu wrote on November 9, 2026 in xata.io's report.

The companies are building a Postgres platform for modern applications and AI agents, pairing production workloads on local NVMe storage with instant database provisioning, instant copy-on-write branching, and scale-to-zero compute. ClickHouse launched ClickHouse Managed Postgres earlier this year, made it generally available the same day, and Xata will migrate its customers there eventually, with no short-term change and sign-ups still open.

This is for Xata customers, ClickHouse Managed Postgres customers, and teams new to either product. The piece follows the branching product, long-lived versus short-lived databases, OLTP and OLAP, a branch per pull request and agent run, and the storage engine and migration plan. The post also points to a ClickHouse announcement.

01Three-second branches on terabyte data

Sarbu writes that agents and software factories now do the grunt work, and that higher coding output has made database workflows the bottleneck. Every pull request and every agent run needs a safe, isolated database with production-like data. Traditional copies are slow and expensive at multiple terabytes, so one database per pull request or agent run does not scale, while copy-on-write branching opens a branch with production data in seconds and does not copy the data.

A Xata branch still takes only around 3 seconds when the database holds multiple terabytes, and the schema and data appear on the child as if copied instantly. The child is fully independent, with its own compute on a different physical node, and nothing there impacts the parent. The post's diagram is labeled 3 active and lists Postgres V18 rows: Production Postgres, cursor/migrate-orders-3c1e, claude/eval-run-7f3a, and maria/add-index Active, and codex/tune-query-2b9e Hibernated.

Instant provisioning creates a fully functional Postgres database, and customers pay for the diff, so 50 branches of a 1TB database cost closer to 1TB than 50TB. Idle databases suspend and wake on the next connection in about one second, and anonymized production data removes sensitive information before leaving production. Sapien, Enginy, David AI, Runner AI, and LoopAI create thousands of branches a week, per pull request, agent run, or SQL session, beside open-source pgstream for logical replication with anonymization and DDL support and pgroll for zero-downtime, reversible schema changes.

02Long-lived Postgres, short-lived branches

Built on local NVMe storage, ClickHouse Managed Postgres is, in Xata's words, the fastest Postgres service available, and not by a small margin. It targets high-throughput, long-lived production workloads, while Xata provisions many small, short-lived databases in seconds, with production-like data, instant branching, and scale-to-zero compute for agentic workloads. Both run unmodified vanilla Postgres and share a strong focus on reliability and stability, the common foundation for integrating the two quickly.

David AI already uses both products together. It told Xata it needs reliable infrastructure to move fast across large volumes of data, that it moved from Neon to ClickHouse Managed Postgres for reliability and performance, and that "our users immediately noticed how much faster everything felt." Xata branching, it said, fits developers and AI coding agents for parallel build and test, and one platform would simplify infrastructure and leave more time to build.

03Postgres OLTP and ClickHouse OLAP

By almost any measure, the post says, Postgres is the most popular OLTP database and ClickHouse the most popular OLAP database, a pair teams have stitched together for years. ClickHouse's Postgres team is making that link first-class through very fast replication from Postgres into ClickHouse and a unified query layer for applications. Xata has wanted that integrated transactional and analytical platform since its early days, when its first product paired Postgres with a search and analytics engine.

Sarbu writes that Xata has long admired ClickHouse and already knew the founders and many of the engineers. That relationship is the stated reason the combined group can take the vision much further. The vision is one platform for transactions and analytics, rather than a setup each team assembles on its own.

04Branch per pull request and agent

Xata's GitHub integration automatically creates an isolated database branch for every pull request, and those branches plug into CI/CD so changes are tested before production. Test branches often scale to zero after only a few minutes, which limits cost because they are not left running. Coding agents follow the same rule: each agent run gets an isolated database and can develop and test without affecting production.

AI platforms that embed Postgres are moving from thousands of databases a week to hundreds of thousands and eventually millions. One use is instant provisioning for every new application, and the other is a branch for every deployment preview so changes are tested against production data. With scale-to-zero, a platform can hold a large fleet of small, mostly idle databases at a fraction of the cost and still serve vanilla Postgres.

05Storage engine, then customer migration

Fast branches and short wake-ups come from the Xatastor storage engine and from warm pools of compute instances. Xatastor separates storage and compute and adds high multi-tenancy, copy-on-write branching, compression, and thin provisioning. It exposes ZFS zvols over the network with NVMe/TCP, and because it works at the block-device level the service runs unmodified vanilla PostgreSQL.

Warm pools remove node and pod provisioning from the hot path of a new database or branch, which speeds provisioning, wake-up after scale-to-zero, and branching. Xata has spent five years on these difficult engineering challenges, and Sarbu says customers create exponentially more branches as usage expands. The team is joining ClickHouse, which Sarbu calls the best position for a vision that is more relevant than ever.

Xata customers can continue as usual while features land in ClickHouse Managed Postgres. After feature parity, the migration is planned to be as smooth as possible, ideally with no impact on the service. ClickHouse Managed Postgres customers can expect branching, scale-to-zero, and an agentic tier, and new users can start on that service or sign up for Xata today.