Aegis DQ

Aegis DQ is a free, Apache 2.0 open-source agentic data quality framework that turns business documentation into executable warehouse validation rules and AI-diagnosed failure reports for data engineers, analytics engineers, compliance teams, and AI-agent builders who need auditable checks across modern data warehouses.

Aegis DQ focuses on a specific gap in traditional data quality tooling: moving from “a rule failed” to “why it failed and how to fix it.” Users point Aegis at business docs such as policies, schema definitions, SLAs, and regulatory requirements, then run aegis generate to produce executable checks. The website states that Aegis can generate notnull, acceptedvalues, and complex custom_sql rules, including CTEs, window functions, and multi-table JOINs. Rules can then run through pipeline manifests that capture the database, rules, docs, LLM configuration, and goal in one reusable pipeline.yaml file. That same manifest can be run from the CLI, Airflow, GitHub Actions, or an MCP client.

The tool’s AI layer is its main differentiator. When checks fail, Aegis classifies severity and returns a plain-English diagnosis, root cause, remediation SQL, and an audit trail of LLM decisions with cost and latency. The provided website includes a real-world AML/fraud demo using 12 BSA/OFAC policies, 6 database tables, and 55 generated rules. In that demo, Aegis detected 11 violations covering CTR, OFAC, SAR, and structuring scenarios, with 5 CRITICAL violations shown in the Hermes example and a total Claude Haiku diagnosis cost of $0.01. The project also advertises 31 rule types and a 5-node LangGraph architecture covering plan, execute, reconcile, remediate, and report.

Aegis DQ is especially relevant for regulated environments where data checks must map back to policies, SLAs, or compliance obligations. It supports six warehouse targets listed on the website: DuckDB, Postgres, Redshift, BigQuery, Databricks, and Athena.