Memon Labs
Book a call

01/LuthorSQL/BI and analytics products

Database questions, answered before they reach the LLM.

People ask your product questions in plain English. LuthorSQL writes the SQL for the ones it is sure about, in under a second, on one GPU.

Book a 15-minute callWatch the sample demo→

ask  Sales by region last quarter?

SELECT region, SUM(amount)

FROM orders

WHERE quarter = 'Q3'

GROUP BY region

sure: 0.940.84 s, no API call

An illustration with invented data.

In plain words

Your users type a question like "what were sales by region last quarter?" Today an LLM turns that into a database query, and you pay for every one.

LuthorSQL can write many of them. In our test it answered 45% by itself. It runs on a machine you own, scores how sure it is, and hands a question to the LLM only when it is not sure.

How it works

  1. 01

    The question comes in.

    From your product, exactly as it does now.

  2. 02

    LuthorSQL writes up to 32 queries and runs them.

    It picks the best one and gives it a confidence score. This takes under a second.

  3. 03

    Sure? It answers. Not sure? Your LLM does.

    The bar is set in advance on your own data. In our registered test, accuracy stayed within half a point of the LLM alone.

LuthorSQL is not an LLM.

It is a System 1 model for databases. An LLM writes its answer one token at a time. LuthorSQL is a 7B model that reads the question once and fills in a typed query plan in a single pass. It runs its best candidates against your database, and a second adapter scores how sure it is. All of that takes under a second on one GPU.

The design was inspired by Jev, TypeSafe's System One model, which returns typed decisions with probabilities instead of text. TypeSafe on System One models →

One question, two ways

The questionWhich five customers spent the most in 2025?

LuthorSQL0.84 s

Fills the whole plan in one pass, in 125 milliseconds, then checks it.

  • showcustomers.name.98
  • showsum(orders.amount).97
  • filteryear(orders.order_date) = 2025.95
  • groupcustomers.id.98
  • ordersum(orders.amount) desc.96
  • limit5.99

It points at values and never writes one. A compiler writes the SQL.

30 candidate queries, run on your database

A second adapter scores each onebar 72%

94% sure. Answered on your servers. No API call.

An LLM2.54 s

Writes the query one token at a time.

SELECT c.name, SUM(o.amount) AS spent
FROM customers c
JOIN orders o ON o.customer_id = c.id
WHERE EXTRACT(YEAR FROM o.order_date) = 2025
GROUP BY c.id, c.name
ORDER BY spent DESC
LIMIT 5;

54 tokens, one after another. Same answer, and you pay for the call.

A sketch, not a recording. LuthorSQL's side plays at its measured pace on one H100. The plan takes 125 milliseconds and the checked answer 0.84 s at the median. The queries, the scores and the LLM's pace are examples.

What you get

  • LuthorSQL, trained on your schema and your kinds of questions
  • A confidence bar tuned on your data before launch
  • Works with Snowflake, BigQuery, PostgreSQL, MySQL, DuckDB and SQLite
  • Fits in about 9 GB, so it runs on one GPU or a Mac Studio-class machine

The honest part

In our registered test LuthorSQL answered 102 of 227 new questions and the bill fell 33.3% at equal accuracy. Your share depends on how hard your questions are. We measure it on your traffic first.

See what LuthorSQL or LuthorText could take off your bill.

Fifteen minutes. Bring a rough idea of what you spend on LLM calls each month.

Book a 15-minute call