aegis-dq
Agentic data quality MCP server — runs structured validation rules against warehouses (DuckDB, BigQuery, Athena, Databricks, Postgres), diagnoses failures with LLM root cause analysis, and proposes SQL remediations. Full audit trail of every AI decision.
README
Aegis DQ
The open-source agentic data quality framework. Validate data contracts, diagnose failures with LLM root-cause analysis, and auto-generate SQL remediation — all in a single CI step or Python call.
- 31 rule types — completeness, uniqueness, validity, referential integrity, statistical, ML anomaly detection
- 6 warehouse adapters — DuckDB, Postgres/Redshift, BigQuery, Databricks, AWS Athena, Snowflake
- Pluggable LLMs — Anthropic Claude, OpenAI, Ollama (local), AWS Bedrock
- Agentic pipeline — plan → parallel validation → LLM diagnose → RCA → SQL remediate → report
GitHub Actions — Quick Start
Add a data quality gate to any workflow in under 2 minutes:
# .github/workflows/data-quality.yml
name: Data Quality
on: [push, pull_request]
jobs:
data-quality:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v4
- name: Validate data quality
uses: aegis-dq/aegis-dq@v0.7.0
with:
rules-file: rules.yaml
db: data/warehouse.duckdb
anthropic-api-key: ${{ secrets.ANTHROPIC_API_KEY }}
The step fails the job automatically when any rules fail, blocking broken data from reaching production. Set fail-on-failure: 'false' to report without blocking.
Offline mode (no API key required):
- name: Validate data quality (offline)
uses: aegis-dq/aegis-dq@v0.7.0
with:
rules-file: rules.yaml
db: data/warehouse.duckdb
no-llm: 'true'
Action inputs
| Input | Default | Description |
|---|---|---|
rules-file |
rules.yaml |
Path to rules YAML |
db |
:memory: |
DuckDB file path |
warehouse |
duckdb |
duckdb · postgres · redshift |
pg-dsn |
— | PostgreSQL / Redshift connection DSN |
no-llm |
false |
Skip LLM — free offline validation |
llm |
anthropic |
anthropic · openai · ollama |
llm-model |
(provider default) | Override the default model |
fail-on-failure |
true |
Fail the step when rules fail |
version |
(latest) | Pin a specific aegis-dq version |
anthropic-api-key |
— | Required when llm: anthropic |
openai-api-key |
— | Required when llm: openai |
Action outputs
| Output | Description |
|---|---|
rules-checked |
Total rules evaluated |
passed |
Rules that passed |
failed |
Rules that failed |
pass-rate |
Pass rate as a decimal (e.g. "91.67") |
report-json |
Absolute path to the full JSON report |
Using outputs in downstream steps:
- name: Validate data quality
id: dq
uses: aegis-dq/aegis-dq@v0.7.0
with:
rules-file: rules.yaml
- name: Post summary
run: echo "Pass rate: ${{ steps.dq.outputs.pass-rate }}%"
Demo

╭──────────────────────────────────────────────────────╮
│ Aegis DQ — RetailCo E-commerce Demo │
│ LLM: amazon.nova-pro-v1:0 via AWS Bedrock │
╰──────────────────────────────────────────────────────╯
✓ Pipeline complete in 7.1s · 12 rules · $0.0056 LLM cost
╭──────────────── Validation Summary ─────────────────╮
│ Rules checked │ 12 │
│ Passed │ 1 │ Failed │ 11 │
│ Pass rate │ 8% │ Cost │ $0.005576 │
╰─────────────────────────────────────────────────────╯
LLM Diagnoses
orders_customer_fk → Order placed with customer_id=99 that does not exist.
Likely cause: customer deleted or test record not cleaned up.
products_sku_unique → Duplicate SKU-001 — two products share the same identifier.
Likely cause: duplicate import from supplier feed.
Remediation SQL (LLM-generated)
orders_status_valid UPDATE orders SET status = 'SHIPPED' WHERE status = 'DISPATCHED';
products_price_positive UPDATE products SET price = ABS(price) WHERE price < 0;
products_stock_non_negative UPDATE products SET stock_quantity = 0 WHERE stock_quantity < 0;
Why Aegis?
| Aegis DQ | Great Expectations / Soda | Monte Carlo / Anomalo | |
|---|---|---|---|
| Open source | ✅ Apache 2.0 | ✅ | ❌ Commercial |
| Agentic LLM diagnosis + RCA | ✅ | ❌ | ✅ Proprietary |
| SQL auto-fix proposals | ✅ | ❌ | ❌ |
| Audit trail (per-decision log) | ✅ | Partial | ✅ Proprietary |
| Pluggable LLM (Anthropic, OpenAI, Bedrock, Ollama) | ✅ | ❌ | ❌ |
| dbt integration | ✅ | ✅ | Partial |
| Portable open rule standard | ✅ | Partial | ❌ |
| ML anomaly detection | ✅ built-in | ❌ | ✅ Proprietary |
Install
pip install aegis-dq
| Extra | What it adds |
|---|---|
aegis-dq[bigquery] |
BigQuery adapter |
aegis-dq[databricks] |
Databricks adapter |
aegis-dq[athena] |
AWS Athena adapter |
aegis-dq[postgres] |
PostgreSQL / Redshift adapter |
aegis-dq[snowflake] |
Snowflake adapter |
aegis-dq[rest] |
REST API server (FastAPI + uvicorn) |
aegis-dq[openai] |
OpenAI LLM provider |
aegis-dq[airflow] |
Airflow AegisOperator |
aegis-dq[mcp] |
MCP server for Claude Desktop |
aegis-dq[ml] |
scikit-learn anomaly detection |
5-minute quickstart
Seed a demo DuckDB database:
import duckdb
con = duckdb.connect("demo.db")
con.execute("""
CREATE TABLE orders AS
SELECT i AS order_id, 'placed' AS status, i * 9.99 AS revenue
FROM range(1, 10001) t(i)
""")
# introduce some bad data
con.execute("UPDATE orders SET order_id = NULL WHERE order_id % 200 = 0")
con.execute("UPDATE orders SET revenue = -5.00 WHERE order_id % 500 = 0")
con.close()
Generate a starter rules file and run:
aegis init
export ANTHROPIC_API_KEY=sk-ant-...
aegis run rules.yaml --db demo.db
Run without an API key (validation only, no LLM diagnosis):
aegis run rules.yaml --db demo.db --no-llm
Pipeline
Every aegis run passes your data through a LangGraph pipeline:
rules (Python / YAML)
│
▼
plan ──► parallel_table ──► reconcile ──► remediate ──► report
│
┌──────────────────┐
│ per table: │
│ execute │
│ classify │
│ diagnose │ ← concurrent across all tables
│ rca │
└──────────────────┘
- plan — parse and validate rules, build an execution graph
- parallel_table — concurrently fans out per table: execute all rules, classify failures by severity, diagnose with LLM, and trace root causes
- reconcile — compare results against expected thresholds
- remediate — LLM proposes a targeted SQL fix for each diagnosed failure
- report — structured JSON + optional Slack notification
Rule types (31 total)
| Category | Types |
|---|---|
| Completeness | not_null not_empty_string null_percentage_below |
| Uniqueness | unique composite_unique duplicate_percentage_below |
| Validity | sql_expression between min_value_check max_value_check regex_match accepted_values not_accepted_values no_future_dates column_exists |
| Referential | foreign_key conditional_not_null |
| Statistical | mean_between stddev_below column_sum_between |
| Timeliness | freshness date_order |
| Volume | row_count row_count_between custom_sql |
| Cross-table | reconcile_row_count reconcile_column_sum reconcile_key_match |
| ML / Anomaly | zscore_outlier isolation_forest learned_threshold |
Example rule:
rules:
- apiVersion: aegis.dev/v1
kind: DataQualityRule
metadata:
id: orders_revenue_non_negative
severity: critical
owner: revenue-team
tags: [revenue, validity]
scope:
warehouse: duckdb
table: orders
logic:
type: sql_expression
expression: "revenue >= 0"
Generate rules with the LLM
Instead of writing rules by hand, let Aegis introspect your table schema and generate a draft rules file:
# Schema-aware structural rules (not_null, between, unique, accepted_values...)
aegis generate orders --db warehouse.duckdb --output orders_rules.yaml
Add a --kb document — any plain text or markdown file describing your business logic — and the LLM generates business validation rules alongside structural ones:
aegis generate orders \
--db warehouse.duckdb \
--kb docs/orders_policy.md \
--output orders_rules.yaml
What goes in a KB file? Anything your team knows about the data:
# orders_policy.md
- status must be one of: placed, confirmed, shipped, delivered, cancelled
- amount must be greater than 0; refunds are handled in a separate table
- customer_id must reference a valid customer (no test accounts: id > 1000)
- order_date must not be in the future
- discount_pct must be between 0 and 0.5 (max 50% discount)
The LLM turns these into accepted_values, sql_expression, between, and foreign_key rules automatically. Generated rules are stamped status: draft — review, promote to active, and commit.
All aegis generate options:
| Flag | Default | Description |
|---|---|---|
--db |
— | DuckDB file for schema introspection |
--kb |
— | Business-context file (text/markdown) |
--output |
rules.yaml |
Output YAML file |
--max-rules |
20 |
Cap on number of rules generated |
--no-verify |
false |
Skip SQL verification of generated rules |
--save-versions |
false |
Persist rules to version store |
--provider |
anthropic |
LLM provider |
--model |
(default) | Override model |
Warehouse adapters
| Adapter | Install | Status |
|---|---|---|
| DuckDB | built-in | ✅ GA |
| BigQuery | aegis-dq[bigquery] |
✅ GA |
| Databricks | aegis-dq[databricks] |
✅ GA |
| AWS Athena | aegis-dq[athena] |
✅ GA |
| Postgres / Redshift | aegis-dq[postgres] |
✅ GA |
| Snowflake | aegis-dq[snowflake] |
✅ GA |
LLM providers
| Provider | Install | Default model |
|---|---|---|
| Anthropic (Claude) | built-in | claude-haiku-4-5 |
| OpenAI | aegis-dq[openai] |
gpt-4o-mini |
| Ollama (local) | aegis-dq[ollama] |
llama3.2 |
| AWS Bedrock | pip install boto3 |
amazon.nova-pro-v1:0 |
Switch providers at the CLI:
aegis run rules.yaml --llm openai --llm-model gpt-4o
aegis run rules.yaml --llm ollama --llm-model llama3.2
aegis run rules.yaml --llm bedrock --llm-model amazon.nova-pro-v1:0
Integrations
| Integration | What it does |
|---|---|
| GitHub Action | CI/CD gate — fails the job when rules fail |
aegis-dq[rest] |
REST API server — aegis serve |
aegis-dq[airflow] |
AegisOperator — drop-in Airflow task |
aegis-dq[mcp] |
MCP server for Claude Desktop / tool use |
aegis dbt generate |
Convert dbt manifest.json to Aegis rules |
CLI reference
| Command | Description |
|---|---|
aegis init |
Generate a starter rules.yaml |
aegis validate <config> |
Check YAML syntax + schema (no warehouse needed) |
aegis generate <table> |
LLM-generate rules from table schema |
aegis run <config> |
Run validation, diagnose failures, produce a report |
aegis rules list |
Browse built-in rule templates |
aegis audit trajectory <run-id> |
Inspect the LLM decision trail for a past run |
aegis audit search <query> |
Full-text search across audit logs |
aegis dbt generate <manifest> |
Convert a dbt manifest to Aegis rules |
aegis mcp serve |
Start the MCP server for Claude Desktop |
aegis run flags:
| Flag | Default | Description |
|---|---|---|
--db |
:memory: |
DuckDB file path |
--llm |
anthropic |
LLM provider |
--llm-model |
(provider default) | Override model name |
--no-llm |
false |
Skip LLM diagnosis entirely |
--output-json |
(none) | Write full JSON report to file |
--notify |
(none) | Slack webhook URL |
--notify-on |
failures |
When to notify: all · failures · critical |
Roadmap
| Phase | Version | Items | Status |
|---|---|---|---|
| Foundation | v0.1 | Core agent, DuckDB, CLI, audit trail | ✅ Done |
| Differentiate | v0.5 | BigQuery, Databricks, Athena, Airflow, Ollama, RCA, ShareGPT export, FTS5 search, dbt, MCP | ✅ Done |
| Quality | v0.7 | SQL verification pipeline, rule versioning, aegis generate (LLM + KB), GitHub Action, ML anomaly detection |
✅ Done |
| Mature | v1.0 | Postgres, REST API, parallel subagents, VS Code extension, eval suite, banking/healthcare packs | 🚧 In progress |
Full issue tracker: github.com/aegis-dq/aegis-dq/issues
Contributing
Contributions are welcome. See CONTRIBUTING.md to get started.
Good first issues: label:good first issue
License
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.