what is dq-copilot-ai-data-quality fr?

DQ Copilot is an open source data quality platform aimed at teams who need to keep the data in their databases and files clean and trustworthy. Other tools in this space, like Great Expectations, Soda Core, and dbt tests, are good at running the checks once someone has written them, but the README argues that writing and maintaining those checks is actually the hard part. Where do new rules come from. Why did a particular row fail a check. Which old rules are still useful, and which have quietly become dead weight. DQ Copilot tries to answer those questions by adding an AI layer on top of the usual rule checking. In practice, this means you can describe a business rule in plain English, such as a definition of what counts as a valid customer status, and an AI agent turns that into an actual technical rule, which a second AI agent reviews before it goes live. When a rule finds a problem, another AI agent looks at the history of past cases to suggest what likely caused it and how it was fixed before. A fourth agent periodically looks across all the rules and recent findings to flag rules that are duplicates, no longer useful, or that leave gaps in coverage. The checks themselves can run against SQL databases, compare two data sources against each other, look for unusual patterns or outliers, catch when a table's structure changes unexpectedly, or even use an AI model to judge softer, less rule-like cases such as whether a product description matches its category. The project can run without connecting to any AI provider at all, in a detection-only mode using its built-in checks. Turning on the AI features requires configuring a provider, and the project defaults to a local-first setup, using a locally run DeepSeek or Ollama model and a local vector database, with external network access turned off by default for privacy. Getting started involves cloning the repository, setting up a Python virtual environment, and running an install script that also handles pulling the local AI model if needed. A dashboard then becomes available in the browser for configuring providers and reviewing results. The project is licensed under MIT and is still in an early, alpha stage. The full README is longer than what was shown.