A data warehouse your business runs without hiring a data engineer.
OpenPlane installs an open Iceberg lakehouse into your own cloud account — AWS, Azure, GCP, Alibaba, Tencent, or a rack in Frankfurt — then models, answers and operates it for you. You pay your cloud directly, so it costs about a quarter of the closed alternatives.
No credit card · no sales call · your data never leaves your account
$ openplane init --cloud alicloud --region ap-southeast-1 ✓ cloud driver alicloud (certified) ✓ kubernetes ACK, 3 nodes, spot ✓ object store OSS, s3-compatible ✓ catalog lakekeeper 0.9 ✓ engine duckdb 1.4 (single node) ✓ conformance 6/6 capabilities passed lakehouse ready in 17m 42s https://acme.openplane.io # nothing was sent to us but metadata.
# the Model agent opens a PR against your repo # models/revenue.yml metric: revenue label: Net revenue expression: sum(invoice.total) filters: [ invoice.status = 'paid', account.is_internal = false ] grain: invoice.issued_at tests: - not_null - non_negative ? Two fields look like revenue. Which one do you report to the board? → invoice.total (selected) invoice.gross_amt
> why did margin drop in August? Gross margin fell 4.2pp to 31.8%. Driver: freight cost per order +38% concentrated in 2 DCs. metric gross_margin defined models/margin.yml:14 lineage orders → shipments → gl_freight scanned 1.4 GB · 2.1s · 0.3 credits [ open definition ] [ show lineage ] # grounded in your metrics, not guessed SQL.
Two crowded markets. Nobody built the bridge.
If you have a data platform team, you are spoiled for choice. If you don't, your options are a closed platform priced for an enterprise, or a spreadsheet.
Open lakehouse vendors
Excellent engines and genuinely open formats — sold to a staff data engineer. The product assumes a platform team already exists.
→ Wrong buyer for the mid-market“Data team in a box” SaaS
Sold to companies with no data team, but as closed SaaS on one or two hyperscalers — with your data in the vendor's account and their margin on every byte.
→ You can never capture the cost savingOpen, any cloud, operated for you
An Iceberg lakehouse a 200-person company can actually run — on the cheapest cloud it can reach, with nobody hired to keep it alive. That's OpenPlane.
→ Currently empty. That's the point.A quarter of the price, because we don't sell you compute.
Closed platforms mark up the machines they run your queries on. We don't run them — you do, in your own account, at your own rate. Most of the saving is structural, not a discount.
A 200-person company
- 12 data sources
- 2 TB lake
- 180 modelled tables
- 40 dashboards
- 60 business users
- ~200 GB processed daily
Monthly, USD. Built from published list rates in September 2026 and our own pricing, including the cloud bill you pay directly. Your numbers will differ — the calculator shows its working.
Your data stays in your account. We only ever see metadata.
OpenPlane splits in two. A small control plane we operate, and a data plane that lives entirely inside your cloud — which is what makes residency, cost and self-hosting all work out at once.
Four agents, each doing a job you'd otherwise recruit for.
Everything they produce is a plain-text file in a git repo you own — reviewable, testable, and yours to take elsewhere. That is the difference between an agent you can trust and a chatbot writing SQL at your warehouse.
OAuth into a source and it infers the schema, the primary and incremental keys, whether CDC is available, and how often to sync from how often the data actually changes. “Extraction strategy” is never a question you are asked.
Proposes your semantic model — entities, metrics, grain, joins — as a pull request you review in business language, not SQL. Ships with starter models for retail, subscription, marketplace and wholesale, so the first draft is mostly right. Ambiguity becomes a question, never a guess.
Conversation and dashboards grounded only in metrics you've defined. Every answer shows the definition it used and the lineage behind it, so your CFO can check the number instead of trusting it. If it can't be answered from defined metrics, it doesn't improvise.
Schema drift, failed syncs, late data, cost spikes, backfills after an upstream fix. Reversible things are handled silently; the rest arrives as “3 things need your attention” with a proposed fix and one button — never as a log file.
Would rather read the source than the marketing?
Fair. The whole data plane is Apache-2.0, the install is a Helm chart, and the conformance suite that decides which clouds we claim to support runs in public CI.
No invented formats. No licence you'll regret depending on.
Every component below is Apache-2.0 or MIT — no BSL, no SSPL, no Elastic Licence anywhere in the dependency graph. Your tables are plain Iceberg on plain object storage: if you leave, they still work.
| Layer | What we use | Licence | Why this one |
|---|---|---|---|
| Table format | Apache Iceberg v3 | Apache-2.0 | Vendor-neutral and production-ready. Not Delta, which one of our competitors steers. |
| Catalog | Apache Polaris · Lakekeeper | Apache-2.0 | Open Iceberg REST catalog. Lakekeeper's small footprint runs the free tier. |
| Engine · small | DuckDB | MIT | Most mid-market queries fit in one machine's memory. This is where the saving comes from. |
| Engine · concurrent | Trino | Apache-2.0 | Federation and BI concurrency when many people query at once. |
| Engine · heavy | Spark + DataFusion Comet | Apache-2.0 | Large batch and ML, vectorised. Only provisioned when you actually need it. |
| Ingestion | dlt · Sling · Debezium | Apache-2.0 | Libraries, not a platform to fight. Real CDC where the source supports it. |
| Transformation | SQLMesh | Apache-2.0 | Incremental by default and column-level lineage native. Bring an existing dbt project and it keeps working. |
| Orchestration | Dagster · Temporal | Apache-2.0 / MIT | Assets, not tasks — the same way you already think about your tables. |
| Semantic layer | OpenPlane spec + SQLGlot | Apache-2.0 | Ours, and open. One metric definition compiles to every engine above. |
| Dashboards | Rill · Evidence · Superset | Apache-2.0 / MIT | Fast exploration, code-first reports, and the self-service long tail. |
We publish which clouds work, and let CI decide.
OpenPlane needs six things from a cloud. A suite tests all six against a live account every night, and the result — not our marketing — is what sets a provider's tier. Certified clouds get a one-click install and a paid SLA.
| Provider | Tier | Object | K8s | Postgres | Secrets | Workload ID | Ingress | Cheapest region |
|---|---|---|---|---|---|---|---|---|
| Amazon Web Services | Certified | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ap-southeast-1 |
| Google Cloud | Certified | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | asia-southeast1 |
| Alibaba Cloud | Certified | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ap-southeast-1 |
| Microsoft Azure | Supported | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | southeastasia |
| Tencent Cloud | Supported | ✓ | ✓ | ✓ | ✓ | partial | ✓ | ap-bangkok |
| Oracle Cloud | Supported | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ap-singapore-1 |
| Hetzner + Cloudflare R2 | Community | ✓ | k3s | CNPG | sealed | keys | ✓ | fsn1 |
| Huawei · OVH · Scaleway | Community | ✓ | ✓ | partial | partial | keys | ✓ | varies |
Community tier means a driver someone else wrote and tests — installable, no SLA from us. Missing your provider? The driver interface is six functions and the conformance suite tells you what's left.
We charge for the work we do, not the machines you own.
No per-seat licence, because we want your whole company using this. No per-row fee, because we shouldn't profit from your growth. No minimum commitment, because that's the thing keeping you off the enterprise platforms in the first place.
Active data products
A monthly fee for each modelled table or metric we keep fresh. You're paying for a maintained data product — the work you'd otherwise hire for.
- Unlimited users and dashboards
- Unlimited storage — it's your bucket
- Price per model falls as you add more
Usage credits
Syncs, materialisations, agent actions and served queries. Row volumes are bracketed, not linear, so a busy month doesn't produce a shock.
- Efficiency rebate: when our optimiser cuts bytes scanned, you keep most of the saving
- Generous free query allowance, so dashboards never feel metered
Verified savings share
We compute what your workload would have cost on Snowflake or Databricks and charge a capped share of the verified difference. If we don't beat them, you don't pay us.
- The benchmark formula is published and reproducible
- Capped, so a good month never becomes a bad invoice
- Best for teams migrating off a closed platform
The free tier fits inside what your cloud already gives away
Not a crippled trial — the same software, sized to run on always-free cloud allowances. It costs us nothing, so we don't need to take it away from you. It ends when your data outgrows a single node, which is a real wall rather than a paywall.
What we will never close.
Plenty of infrastructure companies have relicensed the thing you built on. So here is the line, in public, with a date on it: everything a single team needs to run OpenPlane themselves stays Apache-2.0. We charge for running it for thousands of tenants, which a single team never needs.
- The semantic layer spec and its compiler
- All cloud drivers and the conformance suite
- The connector runtime and every engine adapter
- Model, quality and lineage schemas
- The CLI, and a single-tenant install that genuinely works
- The multi-tenant control plane
- Metering and billing
- The cost optimiser's models, trained across the fleet
- Enterprise SSO, SCIM and audit
- Hosted agents, certification and SLAs
A DCO, not a CLA
We don't ask contributors to sign over rights, because that's the paperwork a company does before it relicenses. We've given up the option.
Trademark, not licence terms
Fork the code freely. You just can't call your fork certified — which is how Linux and Kubernetes did it, and it's enough.
Consulting when you want it
Fixed-scope packages — onboarding, domain modelling, migration off a closed platform — or a regional partner. Never a prerequisite for the product working.
Install it on your own cloud in under twenty minutes.
Or join the waitlist and we'll send the install guide, the conformance matrix, and an honest estimate of what your workload would cost — before you give us anything.