Frequently asked questions
What l0l1 does, what it deliberately does not do, and where the boundaries are.
Is l0l1 a text-to-SQL tool?
No. l0l1 does not translate natural language into SQL. It assumes someone — a person, ChatGPT, or an internal tool — already produced the SQL, and it makes that SQL safer to ship by validating it, scanning for PII, and learning approved patterns.
Does l0l1 send my data to an LLM provider?
l0l1 is not local-only by default. If you configure OpenAI or Anthropic for the validation-reasoning step, query text and schema context are sent to that provider. l0l1 anonymizes detected PII literals before this crossing, but the query structure and schema still leave the process. Providers are pluggable.
What does “validate against live schema” mean?
l0l1 introspects the actual database it is connected to — its real tables, columns, and relationships — and checks the query against them. This catches SQL that is locally plausible but wrong against your schema, such as a misspelled column or a join on the wrong key.
Which databases does l0l1 support?
l0l1 connects to PostgreSQL, MySQL, SQLite, and DuckDB. Schema introspection is cached so validation stays fast against large databases.
How does PII detection work?
l0l1 uses Microsoft Presidio plus additional regex rules to detect SSNs, emails, phone numbers, and credit card numbers in prompts and query text. Detected literals are anonymized before queries are stored or forwarded. Detection is best-effort and misses edge cases — it is not a legal guarantee.
Does l0l1 provide differential privacy or encryption?
No. Result rows are exact, not noised — there is no differential privacy. Queries run against the actual warehouse, so there is no homomorphic encryption. l0l1 reduces what raw data crosses each boundary; it does not transform your results mathematically.
Is l0l1 a compliance product?
No. l0l1 is a tool, not a HIPAA, SOC 2, or GDPR product. It does not confer compliance by transitivity. Your compliance posture depends on how you deploy it and what data you give it access to.
How does pattern learning work?
When a query succeeds, l0l1 sanitizes it of literals and stores it as a shape pattern scoped to your workspace. Those patterns are later surfaced as completions, so approved SQL becomes a shared, growing library instead of tribal knowledge.
What interfaces does l0l1 offer?
A Typer-based CLI, a FastAPI REST server, a Jupyter notebook magic, and a VS Code extension built on an LSP server. The same validation, PII scanning, and completions are available across all four.
Is l0l1 open source?
Yes. l0l1 is MIT licensed and developed in the open by Skelf Research. You self-host it; there is no managed service on the critical path.