Soda (data quality and data contracts)

Soda

What it is

“A data quality platform that helps organizations make sure their data can be trusted” (docs). Three use cases: monitor production data with ML-powered observability, define data contracts, and test data earlier in pipelines via CI/CD.

Maker, history

Soda Data, Inc. (Chicago, IL) with a Belgian entity Soda Data NV (Brussels) (soda.io/company, 2026-10-08). Maarten Masschelein is CEO and Founder. 2025-06-09: Soda acquired NannyML, an ML-monitoring startup whose open-source library estimates model performance without labels (Soda blog). Founding year and funding were not found on any page opened.

Editions and deployment

• Soda Core: Python library and CLI (v4 packages on public PyPI, soda-{data source}); GitHub repo sodadata/soda-core, “Data Contracts engine for the modern data stack”, 2,434 stars, pushed 2026-10-07, created 2020-12-14. The LICENSE file on main is the Elastic License 2.0 (GitHub shows NOASSERTION), which is source-available, not an OSI open-source licence, although the README and docs call the packages open source. No observability (that is Soda Cloud).
• Soda Cloud: managed platform with UI and observability. Pricing: vendor site.

Core architecture

Data contracts written in YAML (README: 50+ built-in checks; schema, types, ranges, constraints) run against PostgreSQL, Snowflake, BigQuery, Databricks, DuckDB and more, from orchestrators (Airflow, Dagster, Prefect) or CI. SodaCL (Soda Checks Language) belongs to the v3 documentation line (docs.soda.io/soda-documentation/soda-v3), which still exists beside v4 contracts.

Role in an enterprise AI rollout

• Quality: Soda’s contracts are intended to catch bad data in pipelines (docs use case); effect on embeddings or feature tables is the author’s inference, not sourced.
• Lineage / access / PII / semantics: not core; combine with a catalog (data-catalogs-compared).

AI features as of October 2026

The earlier version said the docs show no AI features; that was wrong. As of 2026-10-07 the docs have a Soda AI section: chat interface, Contract Autopilot (drafts a first contract), Contract Copilot (edits contracts; a BYOK page covers using your own OpenAI key), Soda MCP (soda-mcp, an MCP server wrapping the Soda Cloud public API for Claude, Cursor, Codex, GitHub Copilot; for Soda Cloud customers, does not need Soda AI enabled), and a root cause analysis (RCA) agent run from your own Claude Code via a Soda plugin, stated to be in private preview (opt-in via support). ML-powered observability is in Soda Cloud. GA/preview status of the other items is not labelled on the pages opened.

Integrations

Warehouses, Airflow, Dagster, BI tools, chat tools (docs overview).

Strengths and weaknesses (opinion)

Contract-first and CI-friendly (per its docs); ease of adoption not assessed. Overlaps with great-expectations (more Python-library oriented) and monte-carlo-data (ML monitoring, closed).

Self-learning

• Docs: https://docs.soda.io/ (free; opened). Markdown index https://docs.soda.io/llms.txt. Certifications: none verified.

Sources

https://docs.soda.io/ ; https://docs.soda.io/llms.txt ; https://docs.soda.io/soda-ai.md ; https://docs.soda.io/soda-ai/soda-mcp.md ; https://docs.soda.io/soda-ai/rca-agent.md ; https://api.github.com/repos/sodadata/soda-core ; https://raw.githubusercontent.com/sodadata/soda-core/main/LICENSE and README.md (all opened 2026-10-07). Second pass (2026-10-08): https://soda.io/company ; https://soda.io/blog/soda-acquires-nannyml ; https://soda.io/sitemap_en.xml.

Open items

• Founding year and funding history: not found (soda.io sitemap has no press or funding posts; agy returned nothing). Whether older Soda Core 3.x releases were Apache-2.0 not checked; GA status of Soda AI features; learning resources.