Data sources

Decisionhouse has no data store of its own. It reads candidate data from files or databases, then uses the query engine to build the optimization model.

Register CSV and Parquet files

Use --register-files for the quickest local workflow.

./deql --register-files inventory.csv candidates.deql query.deql

Pass several files as a comma-separated list.

./deql \
  --register-files gpu_pools.csv,workloads.csv \
  pools_candidates.deql \
  pools_query.deql

Each file becomes a table named after its file stem. inventory.csv is registered as inventory.

Connect a database

Use --connection when the run should query an external database directly.

./deql \
  --connection <url> \
  candidates.deql \
  query.deql

The preview supports data in Postgres, DuckDB, and ClickHouse in addition to CSV and Parquet files.

Use a configuration file

deql-config.toml can list tables, database connections, and server defaults. By default, DeQL looks for it in the working directory or beside the installed binary. Use --config <path> to override both locations.

./deql init

With no config, --connection, or --register-files, the CLI asks where the data is and writes a config. deql init starts the same setup directly. Everything it writes is a default that can be edited later.

Choose data for a run

Option	Use it when
--register-files <files>	The model reads local CSV or Parquet files
--connection <url>	The model reads one external database
--config <path>	The run should use a reusable configuration file

A run that specifies --register-files or --connection already names its data source and does not prompt for a configuration.

Note

The same data options work with deql server. Server mode receives the decision query from a Flight SQL client, so it has no positional query file.