How LoginLlama works

How LoginLlama scores a login: the IP address, the device, where it came from, the user's own history and recent failed attempts, added up into a score from 0 to 10 with the reasons attached.

meta
How LoginLlama works

LoginLlama checks each login to your app and tells you how risky it looks, and why. You send it the user, their IP address and user agent, and whether the password was right. It sends back a score from 0 to 10 and a list of reasons.

This post covers what goes into that score, so you know what you're relying on.

What we look at

The IP address. We match it against lists of Tor exit nodes, known VPN servers, hosting networks and addresses with a record of abuse (FireHOL Level 1). A login from a data centre is not a person at home.

Where the login came from. We look up the country and region from a local MaxMind database. Nothing leaves our servers for this.

The user's own history. Have they used this IP address, this network, this country or this device and browser before? A user who always signs in from Leeds on a Mac doesn't trip anything when they do it again. The same user in another country on a new Windows laptop does.

Impossible travel. If the last login was in London an hour ago and this one is in Sydney, nobody flew that fast.

When they log in. A login many hours away from the time the user usually signs in adds a little to the score.

Failed attempts. We count failures across every check, not just this one. More than 10 failures on one account in a few minutes from one IP means someone is guessing. A correct password straight after that is how a takeover looks. One IP failing on more than 10 different accounts means it's trying one password against many users (password spraying).

How it adds up

Each signal has a fixed weight. A Tor exit node or a success after a burst of guesses counts for more than a new browser. We add the weights up to give the score, and anything above 5 counts as risky. The rules are deterministic: the same login against the same history always gets the same score. Each signal that fired comes back as a reason code, such as known_vpn, new_login_location or impossible_travel_detected, so you can show it to the user or put it in a log.

If you turn on alert emails, the user hears about a login that is new to them, or one with a signal serious enough on its own, such as impossible travel.

What you do with it

LoginLlama doesn't decide for you. You choose what to do at each score. Block above 8, ask for a second factor above 5, or just log it. You can also apply stricter rules to admins than to everyone else.

If you want to talk it through, book a call.