CLI Reference

iPDb's shell binary is ipdb — the standard DuckDB CLI (tools/shell/), unmodified, built with the model/LLM extension available whenever the binary was compiled with it enabled. There are no iPDb-specific flags or REPL behavior beyond the SQL statements themselves; everything below is stock DuckDB CLI usage.

Invocation

ipdb                      # in-memory database
ipdb mydata.duckdb         # open/create a persistent database file
ipdb -c "SELECT 1"         # run one statement and exit
ipdb mydata.duckdb < script.sql   # run a script non-interactively

Common flags (see ipdb -help for the full, version-matched list):

Flag	Effect
-c "<sql>"	Run a single statement/command and exit.
-json	Output results as JSON.
-csv	Output results as CSV.
-readonly	Open the database read-only.
-init <file>	Run a file of dot-commands/SQL before the prompt/script starts.
-no-stdin	Don't read commands from stdin.

Dot-commands

The shell's dot-commands (.tables, .schema, .mode, .output, .read, .help, ...) are unchanged from DuckDB — see Output Formats for the ones most relevant to exporting PREDICT/LLM results, or .help in the shell for the complete list.

Table functions

These are the main entry points for inspecting what's registered in the current database.

ipdb_models()

One row per registered model (CREATE MODEL/CREATE TABULAR MODEL/CREATE LLM MODEL/CREATE EMBED MODEL).

Column	Type	Description
database_name	VARCHAR	Database the model belongs to.
database_oid	BIGINT	Database OID.
schema_name	VARCHAR	Schema the model belongs to.
schema_oid	BIGINT	Schema OID.
model_name	VARCHAR	Model name.
model_oid	BIGINT	Model OID.
comment	VARCHAR	Comment, if any.
tags	MAP(VARCHAR, VARCHAR)	Tags, if any.
temporary	BOOLEAN	Whether the model is temporary.
model_type	VARCHAR	TABULAR / LM / GNN / LLM / EMBED.
model_path	VARCHAR	The PATH the model was created with.
rel_name	VARCHAR	Bound table name, for ON TABLE-style models (empty otherwise).
input_set_count	BIGINT	Number of explicit FEATURES (...) columns.
exclude_set_count	BIGINT	Number of FEATURES * EXCLUDE (...) columns.
opt_rel_name	VARCHAR	Second (edges) table name, for GNN models.
opt_set_count	BIGINT	Feature column count for the GNN edges table.
exclude_opt_set_count	BIGINT	Excluded feature column count for the GNN edges table.
output_count	BIGINT	Number of declared OUTPUT (...) columns.
options	MAP(VARCHAR, VARCHAR)	The model's OPTIONS {...}.
sql	VARCHAR	The original CREATE ... MODEL statement.

ipdb_embeddings()

One row per CREATE EMBEDDING.

Column	Type	Description
database_name	VARCHAR	Database the embedding belongs to.
database_oid	BIGINT	Database OID.
schema_name	VARCHAR	Schema the embedding belongs to.
schema_oid	BIGINT	Schema OID.
embedding_name	VARCHAR	Embedding name.
embedding_oid	BIGINT	Embedding OID.
comment	VARCHAR	Comment, if any.
tags	MAP(VARCHAR, VARCHAR)	Tags, if any.
temporary	BOOLEAN	Whether the embedding is temporary.
table_name	VARCHAR	Table the embedding is attached to.
column_name	VARCHAR	Column the embedding covers.
model_catalog	VARCHAR	Catalog of the backing EMBED model.
model_schema	VARCHAR	Schema of the backing EMBED model.
model_name	VARCHAR	Name of the backing EMBED model.
embedding_size	BIGINT	The SIZE given at creation (0 if unspecified).
sql	VARCHAR	The original CREATE EMBEDDING statement.

duckdb_secrets()

Standard DuckDB secrets introspection — lists configured secrets (name, type, whether persisted, etc.), including any used by CREATE LLM MODEL ... SECRET <name>. See DuckDB's secrets documentation for its columns.

See also

• SQL Reference for CREATE MODEL/CREATE LLM MODEL/CREATE EMBEDDING/PREDICT/LLM syntax.
• Settings for every SET option.
• Output Formats for shell output modes and exporting results.