One engine. Priced by the row.

One credit = 10M rows synthesized, built, and measured against one candidate layout, including measured query latency and bytes scanned on AWS Athena. A 100M-row table across four candidate layouts is 40 credits. You set the scale, and Kijun tests up to four candidate layouts per run. The credit formula is published, so you know the charge in advance.

A clean, readable report in the UI: which Iceberg layout won, what it cost, and the trade-offs across query latency and write load. From there you can export the layout structure in the format you want. The tables the run built stay up afterwards, so you can query them yourself in the playground instead of taking the report's word for it.

Two different things, treated differently:

Plan-included credits (Team, Pro) reset every month. No rollover. They are an allowance, not an asset.

Top-up credit packs expire 12 months after purchase, the OpenAI / Anthropic standard. We consume the oldest credits first, and we warn you at 50%, 75%, and 30 days before expiration. No silent expirations.

Yes. Set a monthly spend ceiling in your workspace settings. When you hit it, we hold all runs until next cycle. You'll never get a bill larger than the cap you set.

Reading, not running. Free opens finished benchmark runs on our own sample schemas, so you can see exactly what a report contains, how a losing layout differs from the winner, and what gets exported. What it does not do is run your workload: your schema, your queries and your targets need a trial or a paid plan, because every real run builds a table and costs real compute.

Every invited account begins on the Pro tier with 60 credits and 60 days, whichever runs out first. What a run costs depends on its row count and how many layouts it tests; the formula in Q.01 prices it before you start. Card required, never charged during the trial. After that you drop to the Free tier automatically. You keep every report you generated. There is no self-serve signup yet, so request early access and we'll open a workspace for you.

Commit to an annual credit baseline and save up to 13%; overage runs at the committed rate. Enterprise contracts include a published 7% auto-escalator from year two. Never a surprise, and negotiable if it matters to you.

Kijun is benchmark testing for databases, and Apache Iceberg is the first supported engine. Today, the only one. MongoDB is the next intended engine, on the roadmap and not built yet. We'd rather ship one engine you can trust than five you can't. Design partners can shape what comes next, apply here.

That's on the roadmap, not built yet. Consuming the baseline downstream will be the natural next step: a CI/CD PR check that flags a regression before merge, and Datadog or Grafana comparing today's production latency against the baseline. Kijun Monitor, which would watch a live table against your baseline continuously, is in development; not available, and not billed. We'll ship these as clearly-marked features when they're real, never billed as if they already are.

For benchmark runs, the product you are buying on this page, Kijun never reads from or writes to your storage. If your table already exists you upload a zip of its metadata, which records types, cardinalities, null rates, ranges and row counts but contains no rows. If it doesn't exist yet you describe it instead. Either way Kijun models from the shape of your schema and queries, never the contents. Synthetic data is generated and measured entirely on Kijun's own infrastructure; nothing from your production environment is copied. (Kijun Monitor, when it ships, will read your live tables by explicit opt-in. It is not available today.)

Card on every tier; invoiced annual billing on Team, Pro, and Enterprise. PO / NET-30 / NET-60 available on Enterprise. All prices are in USD and exclude VAT/GST where applicable. We use Stripe Tax, so your bill will reflect your jurisdiction without your finance team having to ask.