Command line

openpipeline [COMMAND] [OPTIONS]

Run openpipeline COMMAND --help for the same information in your terminal.

Every command that reads a project takes an optional PROJECT argument: the folder that holds dbt_project.yml. It defaults to the current folder, or to OPENPIPELINE_PROJECT_DIR if that is set.

Exit codes: 0 on success, 1 when a model or column you named does not exist or a preview fails, and 2 when the project cannot be loaded or an option has a bad value.

Options for sample rows and the SQL dialect can also be set once in the preview: section of openpipeline.yml. A flag wins over an environment variable, and an environment variable wins over the file.

openpipeline demo

Try everything with no setup.

openpipeline demo [--dir PATH] [--host HOST] [--port PORT] [--serve | --no-serve]
                  [--token TEXT] [--read-only]

Copies the bundled jaffle_shop example, files one realistic pending proposal from an imaginary agent (it adds a revenue_by_month model and touches two business rules, with a rationale), then starts the server.

Option	Default	What it does
--dir PATH	a temporary folder	Copy the example here instead. The folder must be empty or not exist yet.
--host	127.0.0.1	Address to listen on. Also OPENPIPELINE_HOST.
--port	8484	Port to listen on. Also OPENPIPELINE_PORT.
--no-serve	off	Only prepare the folder and print its path.
--token, --read-only		Same as for serve. On the default loopback address the demo is open.

openpipeline serve

Start the web UI and the HTTP API on one port.

openpipeline serve [PROJECT] [--host HOST] [--port PORT] [--reload]
                   [--sample-data PATH] [--dialect NAME]
                   [--token TEXT] [--read-only]

Reads are always open. Changes (saving descriptions, accepting proposals, comments, AI calls) depend on the write mode, which is chosen like this:

1. --read-only: every write is refused with 403.
2. --token: writes need Authorization: Bearer <token>.
3. No token, and --host is 127.0.0.1, ::1, or localhost: writes are open.
4. No token, and any other --host: a random token is generated, printed once with a link that opens the UI with it (http://localhost:8484/?token=...), and used.

See Write access and security for the details.

Option	Default	What it does
--host	127.0.0.1	Address to listen on. Also OPENPIPELINE_HOST. Use 0.0.0.0 in a container; writes then need a token.
--port	8484	Port to listen on. Also OPENPIPELINE_PORT.
--reload	off	Restart on code changes. For developing OpenPipeline itself.
--sample-data PATH	sample_data/	Folder of sample CSV or Parquet files, or a .duckdb file, for sample rows. Also OPENPIPELINE_SAMPLE_DATA.
--dialect NAME	none	SQL dialect of the project, such as snowflake. Also OPENPIPELINE_DIALECT.
--token TEXT	none	Require this Bearer token for writes. Also OPENPIPELINE_TOKEN. Never printed.
--read-only	off	Refuse every write. Also OPENPIPELINE_READ_ONLY=1.

openpipeline mcp

Run the MCP server for AI agents.

openpipeline mcp [PROJECT] [--transport stdio|streamable-http]
                 [--host HOST] [--port PORT] [--sample-data PATH] [--dialect NAME]
                 [--token TEXT]

stdio is a local process started by the agent, with no network port. The HTTP transport needs a Bearer token unless it listens on a loopback address; with no --token on another address, one is generated and printed.

Option	Default	What it does
--transport	stdio	stdio for local agents, streamable-http to run as a service.
--host	127.0.0.1	Address for the HTTP transport. Also OPENPIPELINE_HOST.
--port	8485	Port for the HTTP transport. Also OPENPIPELINE_MCP_PORT.
--sample-data, --dialect		Same as for serve.
--token TEXT	none	Bearer token for the HTTP transport. Also OPENPIPELINE_TOKEN.

openpipeline explain

Explain one model in plain English.

openpipeline explain MODEL [PROJECT] [--json] [--ai] [--business]

Option	What it does
--json	Print JSON instead of text.
--business	Use the glossary's business words instead of model and column names. With --json the technical story is included as technical.
--ai	Also ask Claude to explain it. Needs openpipeline[ai] and an API key. See AI explanations.

openpipeline graph

Print the model graph as JSON.

openpipeline graph [PROJECT]

The output has a nodes list and an edges list.

openpipeline lineage

Show where a column comes from and where it is used. See Search and column lineage.

openpipeline lineage MODEL COLUMN [PROJECT] [--json]

Prints the columns it comes from and the columns that use it, nearest first. With --json it prints the same data as the HTTP API. It exits with 1 if the model or column does not exist.

openpipeline sample

Show a few rows of a model, run on local sample data with DuckDB. Needs openpipeline[preview]. See Sample rows.

openpipeline sample MODEL [PROJECT] [--limit N] [--sample-data PATH] [--dialect NAME]

Option	Default	What it does
--limit N	10	Rows to show, from 1 to 200.
--sample-data PATH	sample_data/	Where the sample data is.
--dialect NAME	none	SQL dialect of the project.

openpipeline rules

List business rules, grouped by owner.

openpipeline rules [PROJECT] [--json]

Warnings about the rules file, such as a link to a model that does not exist, are printed to standard error.

openpipeline review

Summarise what changed between two versions of a project, as Markdown for a pull request comment. See PR summaries.

openpipeline review BASE HEAD [-o FILE] [--ai]

Option	What it does
BASE	The project before the change.
HEAD	The project after the change.
-o, --output FILE	Write to a file instead of printing.
--ai	Add an "In plain words" paragraph written by Claude.

openpipeline version

Print the installed version.