Xata
Postgres platform with copy-on-write branching and scale-to-zero
Custom Search API with Serverless Database in Next.js with Xata
Integrates with
Xata is a managed Postgres platform built around copy-on-write branching and scale-to-zero compute, aimed at per-PR and per-agent database workflows. Vanilla PostgreSQL, open-source core (Apache 2.0), plus pgroll for zero-downtime migrations. The original search/spreadsheet product was retired during a 2024-2025 pivot.
What Xata actually is
Xata is not the product our database describes. The tagline in our records — "serverless Postgres with built-in search," spreadsheet UI, vector search — belongs to the original Xata: an Airtable-flavoured data platform launched in 2022 with a REST API instead of SQL. That product is gone. Somewhere in the 2024-2025 window the company rebuilt around vanilla PostgreSQL and copy-on-write branching, open-sourced the core platform under Apache 2.0 in 2026, and retired the old hosted free tier with it. If you evaluated Xata two years ago, you are looking at a different company now.
What exists today is a Postgres platform whose headline features are instant copy-on-write branching and scale-to-zero compute. Branching happens in the storage layer, not through a Postgres fork, so you keep standard PostgreSQL — extensions, drivers, psql, your ORM. A branch of a terabyte-scale database can be created in seconds because only changed blocks are duplicated. Idle branches suspend compute entirely and bill nothing; on the next connection they wake up, targeting about one second on Xata Cloud and around ten seconds on the self-hosted open-source build. The cofounder's framing on the Postgres.FM podcast is the honest use case: a database branch per pull request, or an ephemeral database per agent run. Production workloads run on the platform too, but the differentiation is the developer workflow, not raw hosting.
Under the hood the open-source distribution runs CloudNativePG (Postgres topology, failover, backups) over OpenEBS for the copy-on-write volumes. The hosted service uses Xata's own closed storage engine, Xatastor, over NVMe/TCP. That split is deliberate: the parts that would let someone else stand up a competing multi-tenant service stayed proprietary. Alongside the platform, Xata maintains pgroll (zero-downtime, reversible schema migrations) and pgstream (data anonymization/streaming), both open source and both usable with any Postgres — in our view the most portable value the company produces.
The company itself is Xatabase Inc., founded in 2020 in San Francisco, Series A backed (Redpoint among the investors, $30M in March 2022, roughly $35M disclosed in total), around 25-27 employees. Small team, real funding, and a demonstrated willingness to kill its own product — hold that thought.
Pricing: what you really pay
Xata is billed on resources: instance-hours plus storage. No per-branch, per-developer or per-database fees, which is the point of the whole design — 5,000 agent branches should not cost 5,000 times a database.
Self-serve Xata Cloud instance rates (US East, us-east-1; billed per hour, 730-hour monthly equivalent shown):
- micro (≤2 vCPU, 1 GB): $0.012/hr, $9/month
- small (≤2 vCPU, 2 GB): $0.024/hr, $18/month
- medium (≤2 vCPU, 4 GB): $0.048/hr, $35/month
- large (2 vCPU, 8 GB): $0.096/hr, $70/month
- xlarge (4 vCPU, 16 GB): $0.192/hr, $140/month
- 2xlarge (8 vCPU, 32 GB): $0.384/hr, $280/month
- 4xlarge (16 vCPU, 64 GB): $0.768/hr, $561/month
- 8xlarge (32 vCPU, 128 GB): $1.536/hr, $1,121/month
EU (eu-central-1) runs at 1.15x the base rate — the micro is $0.014/hr or $10/month. Storage is $0.28/GB/month on Xatastor NVMe, single rate across all regions, billed by the minute, no overprovisioning. Scale-to-zero means an idle branch costs $0 compute and keeps its storage. A 14-day free trial runs without a credit card, and onboarding credits can reach $100 (signup, card connection, first branch, and so on).
Beyond self-serve there is BYOC — Xata Managed, deployed in your own AWS, Azure or GCP account, priced as a quote (a percentage of managed cloud spend; you pay the cloud provider directly for compute and storage). BYOC adds 24/7 support with a dedicated solutions engineer, HIPAA-grade anonymization and a BAA, and capacity tiers sized by concurrent branches (Team up to 1,000; larger tiers up to 10,000 and beyond). The open-source platform is free to self-host on your own Kubernetes.
Two observations. First, the transparency is better than almost anyone in this category: a machine-readable pricing.md published for agents, exact instance tables, one storage rate. Second, the headline "starting at $0.012/hr" is real but narrow — a production cluster with replicas runs 24/7 at full instance rate, and storage is billed across the cluster. Run the calculator before you budget.
Performance & infrastructure
Xata runs vanilla PostgreSQL without core modifications, and its own published comparison claims roughly 3,000 TPS and a cost of $561/month against Amazon Aurora's 2,250 TPS and $2,096/month in their test setup. Treat that as vendor-published, not independent — we have not benchmarked Xata ourselves, and we would point you to our own /benchmarks methodology before drawing conclusions from any vendor's charts. The architecture claim underneath it is credible though: compute-storage separation over NVMe-over-Fabrics with minimal added latency, storage that grows automatically, and copy-on-write branching that avoids full copies.
Region-wise, self-serve currently lists US East (N. Virginia) and EU Central (Frankfurt) on Google Cloud infrastructure; the cloud product advertises multi-region across AWS and GCP, and BYOC runs wherever your AWS/Azure/GCP account does. Compliance: SOC 2 and GDPR on Xata Cloud, plus BAA and advanced anonymization on BYOC.
On reliability specifics we have to be careful: no public uptime SLA that we could find, and no published backup retention policy for Xata Cloud (the open-source stack leans on CloudNativePG's backup management). Failover and replica topology are handled by the platform, but the guarantees are contract-level on BYOC rather than published on the website.
Where it falls short
Track record. Xata has been running this platform for roughly a year or two, after killing the product its existing customers signed up for. The pivot was handled more openly than most — but if you want a database vendor that has run the same architecture for a decade, this is not it.
Then the self-serve gaps: two regions at list price, no published SLA, no stated backup policy on the public pricing page, and the deep compliance and support tiers locked behind a BYOC sales conversation. Multi-tenancy, warm pools and multi-region deployment require Xata Cloud or BYOC — the open-source build covers branching, anonymization and migrations but not the operational extras.
Also note the closed core: Xatastor and the multi-tenant control plane are proprietary, so the open-source version is enough to run Postgres-as-a-platform for your own org but not a fully free equivalent of the hosted service. And the wake-up target of ~1 second is a target — latency-sensitive connection patterns should test it.
Who should pick this
Our pick for teams whose bottleneck is database environments rather than database capacity: engineering orgs where every PR, migration or agent run wants an isolated Postgres and shared staging has become the queue. The copy-on-write branching without per-branch fees is the strongest version of that offer we have seen priced publicly.
We would also shortlist it for platforms that want to offer Postgres to their own customers (there is an OEM program, BYOC or pay-as-you-go), and for teams already invested in pgroll for zero-downtime migrations who want the tool and the platform to come from the same shop.
We would skip it for: enterprises that need a published five-nines SLA and a decade of operational history, workloads that must live in APAC regions at self-serve prices, anyone who wants the spreadsheet-UI/search product (it does not exist anymore), and buyers who need backups-by-policy spelled out on a pricing page.
Key takeaways
- Xata pivoted hard: the "serverless database with search and a spreadsheet UI" is gone; today it is vanilla Postgres with copy-on-write branching and scale-to-zero, with an Apache 2.0 open-source core.
- Branching is the differentiator — instant branches at the storage layer, no per-branch fees; idle branches bill $0 compute and wake in about a second.
- Pricing is unusually transparent: $0.012/hr for a micro (2 vCPU, 1 GB) up to $1.536/hr for 32 vCPU/128 GB, storage $0.28/GB/month, EU at 1.15x. BYOC is quote-based.
- No permanent free tier — a 14-day trial plus up to $100 in onboarding credits; the real free path is self-hosting the open-source platform.
- San Francisco company (founded 2020, ~$35M raised), small team; young platform with no published uptime SLA or backup policy on the website.
- Best for PR-per-database and agent-per-database workflows; wrong fit if you need long operational track records or APAC self-serve regions.
Sources
- https://xata.io/pricing
- http://xata.io/pricing.md
- https://xata.io/
- https://xata.io/blog/serverless-postgres-platform
- https://dev.to/danieltechlab/xata-open-sourced-its-postgres-platform-it-kept-the-part-that-makes-money-closed-26jl
- https://postgres.fm/episodes/xata/transcript
- https://www.cbinsights.com/company/xata-1
- https://www.redpoint.com/companies/xata
Pros
- Instant copy-on-write branching with no per-branch, per-developer or per-database fees
- Scale-to-zero: idle branches bill $0 compute and wake in about a second
- Vanilla PostgreSQL — standard extensions, drivers and tooling, no fork
- Public pricing is unusually transparent, including a machine-readable pricing.md
- Open-source core (Apache 2.0) plus genuinely useful standalone tools: pgroll, pgstream
Cons
- Young platform behind a company that already retired one product its customers relied on
- No published uptime SLA or backup policy on the public site
- Self-serve limited to two regions (US East, EU Central); deep compliance and support tiers are BYOC sales conversations
- Key pieces (Xatastor, multi-tenant control plane) remain proprietary despite the open-source push
Best For
Teams that need an isolated Postgres per pull request or per AI agent run, with branching and scale-to-zero economics rather than per-database pricing
Pricing
Free
- Core features included
Compare with alternatives:
Reviews (0)
No reviews yet. Be the first to share your experience!
Sign in to leave a review.
Sign In →Alternatives to Xata
Upstash
Serverless Redis, Kafka, QStash — per-request pricing
TigerData
PostgreSQL-native time-series database for metrics, events, and analytics
CockroachDB
Distributed SQL database that survives failures
PlanetScale
Serverless MySQL with database branching
Neon
Serverless Postgres with branching and scale-to-zero
Stay in the loop
Get weekly updates on the best new hosting and infrastructure providers, deals, and comparisons.
No spam. Unsubscribe anytime.