Your dbt project. A tenth of the warehouse bill.
Thunderduck runs the dbt project you already have — same models, same tests, same repo — on a serverless DuckDB engine that bills per query-second and sizes every run for you. No migration, no pipelines, no idle warehouse. Point it at your bucket and your databases and run dbt build.
Take the demo, pick your $25 card.
Twenty minutes of your time at dbt Summit — or any time after the show — and we'll send you a $25 gift card. You pick which one.
- $25StarbucksCoffee for the conference week
- $25TargetWhatever you actually need
- $25AmazonThe safe default
- One card per company, for a completed 20-minute demo with someone who works with the data stack.
- Sent by email within five business days of the call — you choose the brand on the call.
- While supplies last. We can't send cards where gift-card incentives break your employer's policy — tell us and we'll skip it.
dbt compatibility that isn't a rewrite.
Connect your dbt project on GitHub — repo URL, branch, a subdirectory if it lives in a monorepo, and a personal access token. Refresh pins it to a commit; Run executes dbt on the same engine as every other query, against the same catalogs. Models materialize as catalog tables you can query in Studio the second the run finishes. Edit in your own IDE, push, refresh, run.
- build, run, test, seed, snapshot and compile — with your own --select and --full-refresh flags
- Models materialize as catalog tables, queryable in Studio with no export step
- Every run keeps its commit SHA, logs and dbt artifacts, downloadable from the run history
- The PAT goes straight to Secrets Manager — the database only ever holds a reference, and it never appears in a query or a log
# your project, unchanged dbt build --select tag:nightly # on Thunderduck: same models, # same tests — materialized into # the catalogs you already query
DuckDB economics, without babysitting a DuckDB instance.
A warehouse bills you for a cluster that is warm whether or not anyone is running dbt. Thunderduck runs each model as its own right-sized job on DuckDB and bills the seconds it actually used.
No idle compute
Nothing is kept warm between runs. A nightly dbt build costs the seconds it ran, not the hours the cluster existed.
Cost known before it runs
Schema, index usage and logical size give every query an estimated cost, runtime and resource footprint before execution — not a surprise line on next month's invoice.
Sizing that learns
Execution plans and past runs pick the CPU, RAM, disk and DuckDB memory limit per model. Nobody picks a warehouse size class, so nobody over-provisions one.
No loading step to pay for
Sources stay in your bucket and your databases. A model can join Iceberg or Parquet with live PostgreSQL or MySQL in one statement, so the pipeline that only existed to co-locate data can go.
Want the arithmetic on your own workload? Run the savings calculator.
Run it beside what you have. Decide with numbers.
Connect, don't migrate
Register your buckets and databases as catalogs. Data stays where it is, in open formats — Iceberg and Parquet. Nothing is copied in.
Point dbt at a second target
Run the same project against Thunderduck in parallel with your current warehouse. Same models, same tests — so the comparison is apples to apples.
Compare cost and results
Query history gives you per-run cost, estimate vs. actual, and the full execution plan for every model. Move the workloads where the numbers say to, on your timeline.
One engine behind your models.
Your models, available to agents
A native MCP server exposes the same catalogs — including the tables dbt just built — to Claude, Cursor and your own agents, with every query cost-gated first.
Zero-ETL federation
Join Iceberg and Parquet in S3, GCS or Azure with live PostgreSQL and MySQL in one SQL statement. No copies, no sync lag.
Every run, visible
Per-run cost, estimates vs. actuals, resource allocation vs. usage, and spend by team — with the cost-per-query trend line.
Runs anywhere
The same engine as SaaS, in your own cloud account (BYOC), or fully on-prem with MinIO for regulated and air-gapped environments.
Twenty minutes, your project, real numbers.
Bring a dbt project and your current bill. We'll show the same models running on Thunderduck and what they cost. Then you pick a gift card.
- A walkthrough on your stack, not a canned deck
- An honest answer on what Thunderduck does not do yet
- Straight talk on pricing, BYOC and on-prem
Can't find a slot that works? Send us a note and we'll work around your schedule.
Bring your dbt project to the booth.
Twenty minutes, your models, your numbers — and a $25 card on us.