I've spent 20 years watching data quality problems get detected downstream — after the bad record is already in the warehouse, the dashboard, the report. I built OpenDQV to move that check upstream: validate records against a YAML data contract before they enter the pipeline.
The core idea: a contract defines what valid data looks like for a domain (banking transaction, healthcare patient, logistics shipment). Every write goes through /validate. Bad data is rejected at the source, not discovered three sprints later.
What's in v1.0.0:
- 24 rule types (regex, min/max, date_format, lookup, checksum, cross_field_range, geospatial_bounds, and more)
- 30 industry contracts across banking, healthcare, insurance, logistics, pharma, and 22 other domains
- Contract lifecycle: draft → review → active, with maker-checker approval. Active contracts are immutable — nothing can silently change a rule in production
- MCP server — works with Claude Desktop and Cursor, with a write guardrail so agents can propose contracts but can't activate them without human approval
- Hash-chained audit log with NTP clock sync at startup
- Importers from Great Expectations, dbt, Soda Core, ODCS 3.1, CSVW, OTel, NDC
- Python SDK, CLI, Streamlit governance workbench
- ~208 req/s on a 2017 laptop (single container, 4 workers)
What it isn't: it's not a pipeline monitor (Monte Carlo does that), not a dbt test framework (Soda does that). It's a write-time enforcement layer.
Solo project, done in my spare time using Claude Code and my AI Team of experts. 1,000+ tests. Runs on Linux, macOS, Windows, and Raspberry Pi.
For now is free. And when we start charging will definitely be cheaper. Also, the project is open source and there is good documentation on how to standup your own infra, if you want to go down that path.
I've spent 20 years watching data quality problems get detected downstream — after the bad record is already in the warehouse, the dashboard, the report. I built OpenDQV to move that check upstream: validate records against a YAML data contract before they enter the pipeline.
The core idea: a contract defines what valid data looks like for a domain (banking transaction, healthcare patient, logistics shipment). Every write goes through /validate. Bad data is rejected at the source, not discovered three sprints later.
What's in v1.0.0: - 24 rule types (regex, min/max, date_format, lookup, checksum, cross_field_range, geospatial_bounds, and more) - 30 industry contracts across banking, healthcare, insurance, logistics, pharma, and 22 other domains - Contract lifecycle: draft → review → active, with maker-checker approval. Active contracts are immutable — nothing can silently change a rule in production - MCP server — works with Claude Desktop and Cursor, with a write guardrail so agents can propose contracts but can't activate them without human approval - Hash-chained audit log with NTP clock sync at startup - Importers from Great Expectations, dbt, Soda Core, ODCS 3.1, CSVW, OTel, NDC - Python SDK, CLI, Streamlit governance workbench - ~208 req/s on a 2017 laptop (single container, 4 workers)
What it isn't: it's not a pipeline monitor (Monte Carlo does that), not a dbt test framework (Soda does that). It's a write-time enforcement layer.
Solo project, done in my spare time using Claude Code and my AI Team of experts. 1,000+ tests. Runs on Linux, macOS, Windows, and Raspberry Pi.
Would love some feedback and support to take it to the next level - https://github.com/OpenDQV/OpenDQV
Thanks,
Sunny Sharma - OpenDQV Dev
reply