Resources
Frequently Asked Questions
Common questions about Featuresmith design, security, and usage
Does Featuresmith send my dataset to third-party AI APIs?
No. Featuresmith v0.1.0 does not contain active LLM integrations or run cloud requests. In future AI phases (Phase 2+), provider integration is strictly opt-in and configured entirely via API keys. Furthermore, the AI layer only receives computed, aggregated statistical summaries (never raw data table rows), ensuring high privacy constraints.
How does Featuresmith scale on very large datasets?
Featuresmith is built on top of Polars, utilizing lazy query execution and multi-threaded calculations to compute statistics instantly. On a standard machine, half a million rows are audited in under 3 seconds. To prevent combinatorial issues on very wide tables, Pearson correlations are capped at 100 columns by default.
Can I add custom connectors and rules?
Yes! Custom connectors can be registered in featuresmith.connectors.registry and custom rules can be registered in the rule engine directly. Dynamic plugin autoloading via packaging entry points is scheduled for Phase 6.