PragmaChangeRead the guide

Know when your API behaves differently.

Train PragmaChange on normal requests you provide. It then checks new behavior where traffic reaches your Nginx proxy.

See the pattern. Choose what to allow, check, or block.

Read the setup guide See how it works
Runs locally in Nginx One model per endpoint No cloud call per request
001 / INTRODUCTIONSCROLL TO EXPLORE ↓
✳

Most tools see a request. PragmaChange sees its recent context.

01 / THE IDEA

Simple path.
Clearer signals.

A request can look ordinary on its own. Its timing, frequency, and values can tell a different story when compared with recent activity from the same client.

01 / 03↗

Choose an endpoint

Start with one API route. Select the request fields that matter for its behavior.

02 / 03↗

Learn its normal

Import representative requests or feature vectors, then train a model for that endpoint.

03 / 03↗

Decide in Nginx

The local module evaluates each matching request and returns allow, check, or block.

Explore the concepts

02 / CONTEXT MATTERS

One client’s activity shouldn’t blur into another’s.

Partition keys keep separate histories for separate clients or accounts on the same endpoint. Time windows turn those histories into useful measurements, such as request counts or average amounts.

The model checks those measurements against patterns learned from representative data.

What is a partition key?
A SIMPLE EXAMPLEPOST /orders
RECENT REQUESTSClient A
each bar = one time slice
↗

A burst appears in this client's recent history.

Illustration of separate histories, not a model score or live decision.

03 / THE DECISION

Three outcomes.
Your rules.

You set the thresholds when you create a release. The score says how unusual a request looks to the trained model. It is not the probability of an attack.

↗
01 / FORWARD

Allow

Forward the request to your API with trusted risk metadata.

◇
02 / REVIEW

Check

Forward it with a signal so your backend can decide how to verify it.

×
03 / STOP

Block

Stop the request at Nginx and return HTTP 403.

04 / BUILT TO STAY LOCAL

Decisions happen at your proxy.

Train and release through the management stack. Nginx loads a validated release and evaluates matching requests locally. If the management stack is offline, the active release keeps running.

Read the deployment guide

05 / MAKE IT YOURS

Your application.
Your integration.

Start with a small reference client. Record dataset files, send application events, or build your own integration from the documented formats.

Node.js

JavaScript & TypeScript

Explore the example ↗

Python

Simple server-side collection

Explore the example ↗

C# / .NET

One implementation for .NET 8

Explore the example ↗

Rust

Native application integration

Explore the example ↗

Examples to study and adapt, not finished SDK products. You can create and upload compatible datasets without them. Repository access is currently restricted.

Read the integration guide
Record now. Train later.Private JSONL files and checked manifests →Understand application actions.Optional workflow models and replay →Manage your proxy hosts.Reviewed deployment, verification, and rollback →

HONEST BY DESIGN

Useful signals.
Clear boundaries.

The Nginx module analyzes request-derived behavior for configured HTTP endpoints. The optional behavior service learns from named actions and confirmed outcomes that your application supplies. Reference clients can help you record data; Nginx does not capture it for training.

It helps you act on unusual patterns. It does not replace application rules, authentication, or your existing security controls.

Read the current limits

START WITH THE GUIDE

Get to know
PragmaChange.

Start with Docker, load the native module, and see how to bring it to your own API.

Open the setup guide