Stacks / Analytics
ClickHouse / DuckDB
Analytics you host. ClickHouse is Apache-2.0; DuckDB is MIT — Snowflake without the credit warehouse.
The problem
Snowflake sells compute credits on a warehouse you do not run. You wanted SQL over events or a dump you already have. ClickHouse is Apache-2.0 columnar analytics you can self-host. DuckDB is MIT embedded analytics for a laptop or a single process. Neither is a clone of Time Travel, Snowflake Marketplace, or every semi-structured function.
Replaces
Snowflake
Estimated SaaS saved
$2,400
per year, small team, before lock-in
Start on RailwayMetabase if people need a clicky BI UI on top of the warehouse
Placeholder: ClickHouse server + a volume, or DuckDB in-process if the job is a file. Dashboards can sit on /stacks/metabase-bi. No one-click template in this repo yet — Start on Railway.
The stack
- ClickHouse
Apache-2.0 columnar analytics database
- DuckDB
MIT embedded analytics if the job is a file, not a cluster
Cost table
| You stop paying | Typical / year | Self-host instead |
|---|---|---|
| Snowflake (small team credits) | $2,400 | ClickHouse or DuckDB |
Agent setup prompt
Replace Snowflake with ClickHouse (preferred for a server) or DuckDB (embedded / local). - ClickHouse: persist data on a volume, bind native + HTTP. DuckDB if the warehouse is a parquet/CSV on disk. - Load one table you actually query. Do not claim Time Travel or Snowflake Marketplace. - Metabase/Superset can sit on /stacks/metabase-bi. Write AGENT_PROMPT.md.
Also as markdown: /stacks/clickhouse-warehouse.md
FAQ
- ClickHouse or DuckDB?
- ClickHouse if several people query a live dataset. DuckDB if the job is a laptop, a CI job, or a single Go process reading parquet. Metabase is the UI — /replace/looker.
- Will my Snowflake SQL run unchanged?
- A lot of ANSI SQL will. Warehouse-specific functions will not. Rebuild the three queries people actually run.