datadile 0.2.0

YAML-defined data tests for database query results

datadile

Datadile helps teams make database assumptions explicit, testable, and visible to both humans and coding agents.

Define lightweight data tests in YAML, run them locally with the CLI, and catch unsafe database states before they break application code, migrations, backfills, or downstream workflows.

Datadile Cloud turns those tests into a shared data-assumption layer for your team and agents. It records test runs, shows which assumptions are passing or failing, schedules automated checks, sends alerts, and surfaces anomalies so agents can use real data context when changing code.

Quick Start

Install Datadile:

pip install 'datadile[postgres]'

Install datadile[postgres] for PostgreSQL data sources or datadile[mongodb] for MongoDB data sources.

Add your project config file. Run this from your project or repo root, or use --global to add a user-level config file:

datadile init [--global]

Then fill in datadile.yaml with your database connection details:

default_data_source: main

data_sources:
  main:
    type: postgresql
    host: localhost
    port: 5432
    user: myuser
    database: mydb
    password_env: DATABASE_PASSWORD

Put the password in an environment variable, not in datadile.yaml:

export DATABASE_PASSWORD='your_password_here'

Recommended: install the Datadile coding-agent skill so your agent can write data tests using Datadile's conventions:

datadile install-skill [--global]

The default install target is .agents/skills/. For agents that don't support .agents/skills/ (e.g., Claude Code), pass --agent, for example:

datadile install-skill --agent claude

Then ask your coding agent to create data tests for your project, for example:

Add tests for the key data assumptions in this code.

If you want to try Datadile without a coding agent, create your first data test in smoke.dile.yaml:

# smoke.dile.yaml
tests:
  - name: database_connection_works
    description: Datadile can connect to the configured database and run a read-only query.
    query: |
      select 1 as value
    expect: "= 1"

Run the tests:

datadile test

Datadile prints a results table and a summary such as 1 passed, 0 failed. To save full, untruncated results to JSON, pass --results-file:

datadile test --results-file datadile-results.json

Data Tests

Tests are defined in YAML files named *.dile.yaml. Each test has name, description, query, and expect. severity is optional and defaults to MEDIUM; valid values are LOW, MEDIUM, and HIGH. data_source is optional and references a named source from data_sources; otherwise Datadile uses default_data_source if one is configured. identity is optional and gives Datadile Cloud a stable identifier for the test, even if the file path changes.

Use the *.dile.yaml naming convention and colocate data tests near the application code they protect:

app/orders/orders.py
app/orders/orders.dile.yaml
app/billing/invoices.ts
app/billing/invoices.dile.yaml

For example, a test can set data_source: app_db after app_db is added under data_sources:

tests:
  - name: no_failed_orders
    identity: orders.no_failed_orders
    description: There should be no failed orders today.
    severity: HIGH
    data_source: app_db
    query: |
      select count(*) as failed_orders
      from orders
      where status = 'failed'
        and created_at >= current_date
    expect: "= 0"

  - name: active_plan_ids
    description: Active subscriptions should only use known plan IDs.
    query: |
      select distinct plan_id
      from subscriptions
      where status = 'active'
      order by plan_id
    expect: "= [1, 2, 3]"

expect is a comparison string. Supported operators are =, !=, >, >=, <, and <=.

For one-row, one-column query results, Datadile compares the scalar value. For multi-row, one-column results, it compares a list of values. For wider results, it compares dictionaries or lists of dictionaries.

Use row_count when the query should return inspectable rows but the assertion is about how many rows were returned:

tests:
  - name: failed_orders_are_limited
    description: There should be at most one failed order today, with rows shown on failure.
    query: |
      select id, status, created_at
      from orders
      where status = 'failed'
        and created_at >= current_date
      order by created_at desc
    expect: "row_count <= 1"

Configuration

Datadile looks for a YAML config file in two locations. Local config takes precedence:

1. ./datadile.yaml (current directory)
2. ~/.datadile/datadile.yaml (user-level)

Create a starter config with datadile init, then fill in your values. To write it somewhere else, pass a destination path. Existing files are not overwritten unless you pass --force.

datadile init path/to/datadile.yaml
datadile init --force

For a user-level config instead of a project-local one:

datadile init --global

default_data_source: main

data_sources:
  main:
    type: postgresql
    host: localhost
    port: 5432
    user: myuser
    database: mydb
    password_env: DATABASE_PASSWORD

Put data source passwords in environment variables, not in datadile.yaml:

export DATABASE_PASSWORD='your_password_here'

Usage

datadile test
datadile test <path/to/file.dile.yaml>
datadile test --results-file datadile-results.json

With no path, Datadile recursively discovers only files matching *.dile.yaml from the current directory. Other YAML files, such as docker-compose.yaml, GitHub Actions workflows, Helm values, and OpenAPI specs, are ignored.

Use datadile context to show existing data tests for the same data source and overlapping tables or columns:

datadile context --data-source app_db --table orders
datadile context --data-source app_db --column orders.status
datadile context --data-source app_db --table orders --format json

The context command loads local *.dile.yaml files. If an API key is configured, it also fetches matching uploaded tests from Datadile Cloud to determine what tests are currently passing and failing.

Coding Agent Skill

Datadile includes a bundled coding-agent skill with Datadile-specific guidance. For many projects, the fastest path is to add datadile.yaml, install the skill, and ask your coding agent to write the first *.dile.yaml tests. The skill is optional and not required to run data tests.

Install it into the current working directory with:

datadile install-skill

By default, this writes the skill to .agents/skills/datadile/SKILL.md under the directory where you run the command. The command shows the full destination path and asks for confirmation before writing the file.

To install for a different coding agent, pass --agent, for example:

datadile install-skill --agent claude

To install into the selected agent's user-level skills directory instead, pass --global:

datadile install-skill --global
datadile install-skill --agent claude --global

To install it somewhere else, pass a destination path:

datadile install-skill path/to/SKILL.md

Datadile Cloud

If an API key is configured, Datadile records completed test runs in Datadile Cloud. Datadile monitors runs to alert you about failures and anomalies within your tests. Datadile also draws on this data to provide context to your coding agent, allowing it to take into account what data assumptions are failing while writing your code.

You can schedule tests to run automatically by connecting your data sources to Datadile Cloud, either through encrypted pipelines or within your own private cloud.

To enable cloud features, put the API key in an environment variable and reference that variable from datadile.yaml:

export DATADILE_API_KEY='your_api_key_here'

api_key_env: DATADILE_API_KEY

Data source entries can also reference an ID if the connection details are stored on your Datadile account:

api_key_env: DATADILE_API_KEY

default_data_source: warehouse

data_sources:
  warehouse:
    id: ds_abc123