HYX

Product plane / AI engine

AI engine

operational

Request risk decisions with controlled Hyx AI releases.

Hyx AI reviews requests at the edge. Separate quality checks govern learning, release, and rollback.

01 / proof
~4µs
Hyx AI review
Completed beside the request.
02 / proof
38
decision signals
Request details and recent visitor behavior.
03 / proof
2
release checks
Quality checks before an update goes live.
04 / proof
1 click
global rollback
Return to the previous version immediately.

Core capabilities

Clear decisions with reasons.

Hyx AI returns a confidence level and the reasons behind it. You choose how each site responds, while review mode shows the effect of an update before it changes protection.

AI-01 / feature contract

trainer = edge / v38

Verdict latency

~4µs

Hyx AI makes the decision locally, with no extra service call.

38-slot vector

16 actor / 22 request

0001020304050607080910111213141516171819202122232425262728293031323334353637

Current reasons

01error rate02stored content03asset mix04bot risk05rule risk06client identity07query pattern

Actor context

16

bounded 60-second window

Request shape

22

current request only

Train / serve skew

0

one FEATURE_NAMES contract

01

Edge request scoring

Hyx AI reviews each request on arrival so policies can stop attacks before they reach your site.

02

Versioned feature contract

Hyx AI reviews each visitor using recent behavior and request details.

03

First-request enforcement

High-confidence attacks can be stopped on the first request without a warm-up period.

04

Rolling behavior history

Hyx AI uses recent visitor behavior to identify abuse patterns.

05

Calibrated confidence scores

Calibrated scores keep a 90% result consistent enough to support clear policy thresholds.

06

Automated retraining and rollout

Hyx AI learns from real traffic and releases approved versions across the network automatically.

How it works

Learning and release are separate stages.

Traffic helps Hyx AI improve, but updates do not go live automatically. Independent quality checks and approval steps protect each release.

AI-02 / guarded learning cycle

stage 01 / 05

01

score

Review

Hyx AI considers 38 request and behavior signals before the request reaches your server.

01
engine
Hyx AI
02
output
confidence + reasons
03
latency
~4µs

Hyx AI learns continuously / updates require approval

  1. 01

    Observe

    Review the same set of signals for each request and record the eventual outcome.

  2. 02

    Learn

    Combine confirmed outcomes and team feedback into lessons for Hyx AI.

  3. 03

    Check

    Compare each update with the live version for accuracy and consistency.

  4. 04

    Release

    Publish only after checks pass and keep the previous version ready for rollback.

Operational detail

Capabilities included with Hyx.

A concise list of what this part of Hyx provides. Roadmap items are labelled explicitly.

Open the full matrix

Included

19

  • Edge request scoring

  • Versioned feature contract

  • First-request enforcement

  • Rolling behavior history

  • Calibrated confidence scores

  • Automated retraining and rollout

  • Hyx AI release checks

  • High-confidence training labels

  • Prediction-error feedback

  • Human review queue

  • Decision explanations

  • Configurable enforcement modes

  • Promotion to persistent bans

  • Hyx AI quality monitoring

  • Hyx AI rollback controls

  • Proposes new protections for review

  • Shadow evaluation

  • Single-location canary rollout

    roadmap

  • Network-level attack aggregation

    roadmap

Next system / Observability

Inspect Hyx AI decisions in request records.

Explore request history, decision reasons, analytics, version tracking, and operational alerts.

Continue to Observability