Configuration

OpenPipeline has very little to configure. Everything has a sensible default, and your dbt project stays the source of truth.

Environment variables

Variable	Default	What it does
OPENPIPELINE_PROJECT_DIR	the current folder	Which dbt project to open when you do not pass a path. The Docker image sets it to /project.
OPENPIPELINE_HOST	127.0.0.1	Address the web UI and the MCP HTTP transport listen on. The Docker image sets 0.0.0.0.
OPENPIPELINE_PORT	8484	Port of the web UI and API.
OPENPIPELINE_MCP_PORT	8485	Port of the MCP server with --transport streamable-http.
OPENPIPELINE_SAMPLE_DATA	sample_data/ in the project	Folder of sample files, or a .duckdb file, for sample rows.
OPENPIPELINE_DIALECT	none	SQL dialect of the project, for sample rows. For example snowflake.
OPENPIPELINE_AI	on, if a key is present	Set to off, 0, or false to switch AI explanations off completely.
OPENPIPELINE_AI_MODEL	claude-opus-5-5	Which Anthropic model to use for AI explanations.
ANTHROPIC_API_KEY	none	Your Anthropic key. ANTHROPIC_AUTH_TOKEN and ANTHROPIC_PROFILE also work, and so does a profile in ~/.config/anthropic made with ant auth login.

Command line options win over environment variables, and environment variables win over openpipeline.yml.

openpipeline.yml

The openpipeline.yml file sits next to dbt_project.yml. It has three optional sections: rules: for your business rules, glossary: for your business words, and preview: for sample rows.

preview:
  sample_data: sample_data   # a folder or a .duckdb file, relative to the project
  dialect: snowflake

rules:
  - id: shipped-revenue
    name: Revenue counts once an order is shipped
    description: >
      An order only counts as revenue after it has shipped.
    owner: finance-team
    tags: [finance, orders]
    applies_to:
      - model: orders
      - model: daily_revenue
        columns: [revenue]

Key	What it does
rules	A list of business rules. The fields are explained on the rules file format page.
glossary	A list of terms: name, label, and optionally definition and aliases. See business words.
preview.sample_data	Where sample data lives, relative to the project folder. Same as --sample-data, but set once for everyone.
preview.dialect	SQL dialect of the project. Same as --dialect.

Other keys in preview: are ignored, and so is a preview: section that is not a mapping. The file is optional.

Files OpenPipeline reads and writes

Path	Read	Written
dbt_project.yml, models/**, seeds/**	yes	only YAML description edits you save, or proposals you accept
target/manifest.json	yes, when it exists	never
openpipeline.yml	yes	only when you accept a proposal that changes it
sample_data/	yes	never
.openpipeline/proposals/	yes	when an agent files a proposal, or you comment, accept, or reject
.openpipeline/cache/ai/	yes	when you ask for an AI explanation

Add .openpipeline/ to your .gitignore, unless you want to commit proposals.

OpenPipeline never connects to your warehouse and never runs dbt.