Start small. Prove it. Then scale.
No big licence, no new hardware, no six-month project. Thunderduck runs in a small corner of the Kubernetes cluster you already have, next to the S3-compatible storage you already run. Start small, prove it on real work, then scale.
What it takes to start.
- A Kubernetes cluster. A few nodes are enough to start; Thunderduck installs into its own namespace with Helm.
- S3-compatible object storage. MinIO, RustFS, SeaweedFS, Ceph or any store that speaks the S3 API holds your Iceberg tables and files.
- A PostgreSQL database for the platform's own settings and history.
Kubernetes
MinIO
RustFS
SeaweedFSGet value before you make the big decision.
Most on-prem data platforms ask for the investment up front and deliver the value later. Thunderduck turns that around.
- A large licence and a hardware order before the first query
- Months of installation and tuning
- Sized for peak load on day one, idle most of the time
- A business case to write before anyone has seen it work
- A few nodes in your existing Kubernetes to start
- Installed with Helm charts, running in days
- Query workers start on demand and stop when the query ends
- Value first, then grow the footprint as usage grows
- 1
Start small
Install into one namespace with modest resources. Connect a bucket and a database or two, and give a few people access.
- 2
Prove it
Run the real questions, notebooks and reports. Query history shows who used what and what it cost in CPU and memory.
- 3
Scale
Add nodes and raise the worker sizes when the usage is there. Nothing to migrate: the data stays in your storage.
Everything that makes it useful,
inside your walls.
MCP for your agents
Claude and other agents query your on-prem data through the same MCP server and guardrails as the cloud.
Isolated compute
Every query gets its own worker in your cluster, sized from its plan and stopped when it finishes.
Your network, your rules
Data, credentials and compute stay inside your perimeter. Roles control who can read or write each catalog.
Open formats
Tables are Iceberg and Parquet in your own storage. Nothing proprietary sits between you and your data.
