Feyn Labs launched SQRL, a family of models that turn plain English into SQL.
The models probe a database with read-only checks before committing to a query.
The flagship reports 70.6% execution accuracy on business intelligence tests.
Turning a plain-English question into a working database query sounds simple. In practice, it’s where most AI data tools fall apart. A new release from Feyn Labs takes a smarter approach.
Look First, Then Query
Feyn Labs has released SQRL, a family of text-to-SQL models. The idea behind them fixes a common failure in AI data tools.
Most text-to-SQL systems guess. You ask a question, and the model writes a query based on what it assumes your database looks like. If it guesses the table names or structure wrong, the query fails or, worse, returns confident nonsense.
SQRL works differently:
It inspects the database first with read-only probes
It learns the real structure before writing anything
It then commits to a query based on what actually exists
The flagship, SQRL-35B-A3B, reports 70.6% execution accuracy on BI benchmarks
That “look before you leap” design is the whole point. By checking the database before writing, it avoids the confident-but-wrong answers that make AI data tools risky for real business use.
Why It Matters For SaaS Teams
If your product or team relies on people pulling data, this kind of tool changes who can do it.
The practical upside:
Non-technical staff can ask questions in plain English and get real answers
Analysts save time on routine queries and focus on harder work
Fewer errors because the model verifies structure instead of guessing
Read-only probes mean it inspects safely without changing your data
Text-to-SQL has been a crowded space, and plenty of tools promise this. The difference SQRL is pitching is reliability. A 70.6% execution accuracy figure is meaningful, though it’s worth remembering it’s an early, company-reported number on specific benchmarks, not an independent test across every real-world database.
Two honest caveats before you get excited. First, that accuracy comes from Feyn Labs’ own evaluation, so treat it as a claim until third parties test it. Second, even a strong text-to-SQL tool needs a clean, well-structured database to work well. Point it at a messy schema and no amount of clever probing fully saves you.
Still, the direction is right. Tools that verify before they act, rather than guess and hope, are exactly what business data work needs. If SQRL holds up under independent testing, it’s a useful addition for any team drowning in data requests.
Quick Links:
Comments
Be the first to leave a comment.