Talk to us

From your data to watch around the clock.

Connect your data and describe the behaviour. The agent builds and proves the detector, then Lucir watches day and night.

›Flag fraudulent transactions. Live
590,540rows read
118,107unseen, scored
0.926ROC-AUC
Illustration. The build figures are Lucir's real IEEE-CIS run; the events are examples.

How it works

Four steps. Two of them are yours.

  1. 01

    Connect

    Point Lucir at your database. Detectors run there as queries.

    No data migration.

  2. 02

    Describe

    Say what matters, in plain language or with a policy.

    No rule writing.

  3. 03

    Prove

    The agent builds a detector, checks it, and scores it on history it never saw.

    Evidence before go-live.

  4. 04

    Watch

    You approve it. Lucir runs it around the clock, on the cadence you set.

    Alerts with their reasons.

What changes

The rulebook, rebuilt around evidence.

Detection rules today compared with Lucir
TodayWith Lucir
A new ruleWeeks of tickets and vendor timeMinutes to hours, built by the agent
Who writes itAn engineer translating a requestThe agent, from the request itself
What it is tested onOften the data it was tuned onA locked holdout, never used to tune
What runs in productionWhatever was last editedFrozen at approval, retrained models versioned
When an auditor asksA document describing the ruleThe detector, its scores and its history

Where it runs

Built on your data. Owned by you.

  • Queries run in your database. Snowflake, BigQuery and Databricks support is next.
  • Least exposure. The agent works from profiles and small samples of your data, with held-out labels masked.
  • Your model provider, your keys.
  • Any domain. Nothing is hard-coded to one dataset.

See it build on your data.

Talk to us