Data Quality Studio — Great Expectations & Soda Core

Build production-ready data quality tests visually, no memorising SodaCL or Great Expectations syntax.

Connect to your live database or just point it at local CSV/Excel/Parquet files and browse tables directly in the VS Code sidebar, pick checks from a full catalog filtered to each column's data type, and generate ready-to-run SodaCL YAML or Great Expectations Python in one click.

How It Works

1. Connect and browse your database

The extension auto-detects your credentials from ~/.dbt/profiles.yml or a workspace .env file on startup for database connections. Your schemas and tables appear immediately in the Data Quality sidebar, no configuration required if you already have dbt set up.

No database? Click ⚙ → Local files and select CSV, Excel (.xlsx), or Parquet files instead, each file appears as a table with its columns and inferred types, and the generated tests query the files with DuckDB. Your file selection is remembered across sessions.

Click any table to open the test builder panel.

2. Pick your framework — once per session

Choose Soda Core or Great Expectations the first time you open a table. The choice is remembered across VS Code and you will never be asked again unless you explicitly switch.

Checks are tailored to the framework you pick. There is no mixing: Soda checks produce SodaCL YAML, GE checks produce Python.

3. Add checks per column

Every column is shown with its data type badge. Click a column to expand it and use + Add check to open VS Code's native command palette, searchable, keyboard-navigable, and filtered to checks that apply to that column's type.

Numeric columns get range checks (min, max, avg, sum, percentile). Text columns get length and regex checks. Timestamp columns get freshness and date-range checks. Booleans and categoricals get value-set checks. Every column gets null and uniqueness checks.

4. Generate — opens in a new editor tab

Click Generate Tests. The output opens in a new editor tab with the correct language mode already set (YAML for Soda, Python for GE). Save the file wherever your project expects it, there is no fixed output path.

Features

• Zero-config auto-connect: Reads ~/.dbt/profiles.yml (resolves {{ env_var('...') }} templates), workspace .env, or a custom path you set once in VS Code settings
• Works without a database: Select local CSV, Excel, or Parquet files. Column names and types are inferred from the files themselves, and generated tests run against them with DuckDB (soda-core-duckdb / duckdb-engine). No server, no credentials
• Live schema browser: schemas → tables loaded directly from your database; refresh any time with the ↺ button
• One-time framework choice: Pick Soda Core or Great Expectations once; the choice persists across sessions and across tables
• Full check catalog: 18 Soda checks + 17 GE checks, each filtered to the column data types they apply to:
  • Presence & validity — null, uniqueness, value sets, allowed values
  • Numeric statistics — min, max, average, sum, std dev, percentile, unique proportion
  • String patterns — regex match, length bounds, date format strings
  • Timestamps & freshness — freshness window, date min/max, parseable dates
• VS Code-native check picker: + Add check opens the command palette at the top of the editor, not a clipped inline dropdown
• Custom checks: Write any check the catalog doesn't cover: give it a name and a SQL fail condition (rows matching it fail). The condition runs in your warehouse's own SQL dialect shown in the field label and generates a SodaCL failed-rows check or a GE UnexpectedRowsExpectation depending on your framework
• Generated output uses your real connection details: host, port, database, user, and schema are filled in; only the password stays as an environment variable placeholder
• Framework-specific configuration hints: Soda output includes a ready-to-fill configuration.yml block; GE output includes the correct pip install line and DATABASE_URL export command

Supported Data Sources

Custom credentials path: Set dq-studio.credentialsPath in VS Code Settings to point at any profiles.yml, BigQuery service account JSON, or .env file. The extension uses that path instead of the default lookup. Supports ~ for the home directory.

Local files — how it works

• Each selected file becomes a table (named after the file, e.g. orders.csv → orders) under a files schema in the sidebar
• CSV: column types are inferred from a sample of the file (integer, double, boolean, timestamp, varchar) — the same way DuckDB's read_csv_auto sniffs them. Excel: types are read from the typed cells of the first sheet. Parquet: types come straight from the file's own schema metadata
• Generated Soda output uses a type: duckdb data source pointing at the file (pip install soda-core-duckdb). Excel needs a one-time bootstrap — the generated file includes the exact one-liner that builds a small .duckdb view over the sheet, since Soda cannot open .xlsx directly
• Generated GE output uses an in-memory DuckDB engine with a query asset over read_csv_auto(...) / read_xlsx(...) / read_parquet(...) (pip install great_expectations duckdb duckdb-engine). DuckDB auto-installs its excel extension on first use
• Custom SQL fail conditions work exactly as with a warehouse.

Full Check Catalog

Soda Core (SodaCL)

Great Expectations

Generated Output

Soda Core

# Soda Core — checks file
# Generated by Data Quality Studio
#
# 1. Install:  pip install soda-core-postgres
#
# 2. Save the following as configuration.yml:
#
#   data_sources:
#     raw_data:
#       type: postgres
#       connection:
#         host: prod.db.example.com
#         port: 5432
#         database: warehouse
#       username: analytics_user
#       password: ${DB_PASSWORD}
#       schema: raw_data
#
# 3. Run:
#      soda scan -d raw_data -c configuration.yml checks.yml

checks for raw_data.order_items:
  # order_item_id
  - duplicate_count(order_item_id) = 0
  - missing_count(order_item_id) = 0
  # amount_usd
  - min(amount_usd) >= 0
  - max(amount_usd) <= 99999
  - missing_count(amount_usd) = 0
  # status
  - invalid_count(status) = 0:
      valid values: ['pending', 'shipped', 'delivered', 'returned']
  # custom check
  - failed rows:
      name: Amount cannot be negative
      fail condition: amount_usd < 0

Great Expectations

# Great Expectations (GX Core 1.x) — raw_data.order_items
# Generated by Data Quality Studio
#
# pip install 'great_expectations[postgresql]'
#
# Usage:
#   export DATABASE_URL="postgresql+psycopg2://analytics_user:YOUR_PASSWORD@prod.db.example.com:5432/warehouse"
#   python <this_file>.py

import os
import great_expectations as gx

context = gx.get_context()

# ── Data source ─────────────────────────────────────────────────────────────
# Replace YOUR_PASSWORD, or export DATABASE_URL to skip this line entirely
CONNECTION_STRING = os.environ.get("DATABASE_URL", "postgresql+psycopg2://analytics_user:YOUR_PASSWORD@prod.db.example.com:5432/warehouse")

data_source = context.data_sources.add_sql(
    name="raw_data_datasource",
    connection_string=CONNECTION_STRING,
)
asset = data_source.add_table_asset(
    name="order_items",
    table_name="order_items",
    schema_name="raw_data",
)
batch_definition = asset.add_batch_definition_whole_table("order_items_full_table")

# ── Expectation suite ───────────────────────────────────────────────────────
suite = context.suites.add(gx.ExpectationSuite(name="order_items_suite"))

# ── Checks ──────────────────────────────────────────────────────────────────
# order_item_id
suite.add_expectation(gx.expectations.ExpectColumnValuesToBeUnique(column="order_item_id"))
suite.add_expectation(gx.expectations.ExpectColumnValuesToNotBeNull(column="order_item_id"))

# amount_usd
suite.add_expectation(gx.expectations.ExpectColumnMinToBeBetween(column="amount_usd", min_value=0, max_value=0))
suite.add_expectation(gx.expectations.ExpectColumnMaxToBeBetween(column="amount_usd", min_value=0, max_value=99999))

# amount_usd — custom: Amount cannot be negative (rows matching the condition fail)
suite.add_expectation(gx.expectations.UnexpectedRowsExpectation(
    unexpected_rows_query="SELECT * FROM {batch} WHERE amount_usd < 0",
    description="Amount cannot be negative",
))

# ── Run validation ──────────────────────────────────────────────────────────
validation = context.validation_definitions.add(
    gx.ValidationDefinition(name="order_items_validation", data=batch_definition, suite=suite)
)
results = validation.run()
print(results)

Connection Setup

Auto-detect (no configuration needed)

On startup the extension tries the following in order:

1. The path set in dq-studio.credentialsPath (if configured)
2. ~/.dbt/profiles.yml — reads your active target and resolves {{ env_var('VAR') }} templates against your shell environment
3. A .env file in the open workspace root (looks for DB_HOST, DB_USER, DB_NAME and variants)

If a connection is found, the sidebar tree populates automatically.

Manual connection

Click the ⚙ icon in the Data Quality sidebar title bar. You can:

• Use an auto-detected connection if one was found, confirm or browse for a different file
• Browse for a credentials file: opens a file picker defaulting to ~/.dbt/
  • profiles.yml or .yaml — dbt profile format, all targets supported
  • .env — key=value format with standard DB_HOST / POSTGRES_* / DATABASE_* variable names
• Paste a connection string: postgresql://user:password@host:5432/database
• Select a BigQuery service account JSON: standard {"type": "service_account", "project_id": ...} format

Persistent credentials path

Set dq-studio.credentialsPath in VS Code Settings (or settings.json) to avoid browsing each time:

{
  "dq-studio.credentialsPath": "~/.dbt/profiles.yml"
}

Supports ~ for the home directory. Accepts profiles.yml, service account JSON, or .env files.

Requirements

• VS Code 1.85 or later
• A running PostgreSQL, Redshift, Snowflake, or BigQuery database accessible from your machine — or local CSV/Excel/Parquet files (no database required)
• Python with great_expectations (GX Core 1.x) or soda-core-* installed, only needed to run the generated tests, not to use the extension itself

Privacy

The extension connects only to the database you configure. No data, queries, or schema information is sent anywhere other than your own database server.

License

MIT — see LICENSE