Skip to content

Kysira Documentation

Kysira hunts and kills attacks in milliseconds using AI-powered intent analysis before it can touch your application, data, or customers.


How it works

Every inbound request is scored by a combination of ML models and an advanced processing layer covering the full OWASP Top 10 and Zero days. The score drives one of two actions:

  • Shadow mode — request passes through; score and reason are logged and surfaced in the dashboard. Safe for initial rollout with zero traffic risk.
  • Active mode — requests above the kill threshold are blocked with a 403 before reaching your application.

Deployment models

Kysira integrates at whichever layer makes sense for your infrastructure:

Model Best for
Standalone proxy Any stack — sits in front of your app, no infra changes required
Istio ext_proc Istio mesh — zero extra hop, Envoy sidecar handles inspection
Standalone Envoy Envoy API gateway / Contour — configure the filter in envoy.yaml
nginx auth_request nginx — thin adapter sidecar, no custom nginx build required

Get started

  • New here? Follow the Quickstart — local and minikube paths, under 10 minutes.
  • Deploying into a customer cluster? See the Client deployment guide.
  • Already running Grafana or Datadog? See Observability — Kysira is discoverable with a single values flag.

Components

Component Language Role
kysira-inference Python / FastAPI Scores requests using ML models + regex detectors
kysira-ext-proc Go Envoy external processing (gRPC) integration
kysira-proxy Go Standalone reverse proxy
kysira-nginx-auth-adapter Go nginx auth_request target