Cyber

Trust and data handling

Your data does not go anywhere.

Senua Cyber runs inside your perimeter, on hardware you own. There is no vendor cloud in the request path and no telemetry channel out. This page shows exactly what that means, component by component.

Where your data goes

Shield sits inline. Nothing leaves.

A request arrives, your reverse proxy asks Shield for a verdict, and the request either continues to your application or is refused. Every box below runs on your infrastructure. There is no arrow leaving it, because there is nothing to draw.

Senua Shield request path A visitor request enters your reverse proxy, which consults Shield running on your own host. Shield returns allow or block. Allowed traffic continues to your application. Shield's causal patterns and audit log stay on your disk. No component connects outside your perimeter. Your perimeter Visitor or attacker Reverse proxy nginx / HAProxy ask verdict Senua Shield verdict in ~0.09 ms your process, your host Patterns + audit log your disk, your retention allowed traffic continues Your application unchanged no egress no vendor cloud Senua Shield request path The same request path drawn top to bottom for a narrow screen: a visitor reaches your reverse proxy, the proxy asks Shield on your own host and gets a verdict back, Shield keeps its causal patterns and audit log on your disk, and allowed traffic continues to your application. Nothing leaves your perimeter. Visitor or attacker Your perimeter Reverse proxy nginx / HAProxy ask verdict Senua Shield verdict in ~0.09 ms your process, your host Patterns + audit log your disk, your retention allowed Your application unchanged no egress no vendor cloud

Compare this with a SaaS bot-defence product, where the first hop is the vendor's edge network and your production traffic is decrypted outside your control. That is the architectural difference, not a policy promise. The ~0.09 ms verdict in the diagram is the per-request figure measured on CSIC 2010, a published academic HTTP request corpus.

What is logged

Logging and audit.

Every verdict is recorded with the evidence that produced it, on your disk, under your retention policy. A block you cannot explain is a block you cannot defend to the customer it affected.

RecordContentsWhere it livesRetention
Verdict logSource id, verdict, this request's departure, the source's running mean, requests seenYour diskYou set it
Audit logLicence installs, policy changes, mode changes, operator and timestampYour diskYou set it
Learnt baselineCausal patterns learnt from your own normal trafficYour diskUntil you rebuild it
Request bodiesNot retainedNowhereNot kept
Telemetry to SenuaNone. There is no channel.NowhereNot kept

What you get for your authorisation pack

Built for an Australian security review.

Self-hosted software sits inside your own system boundary, which is the shortest path through a security review there is. Here is what we hand your team for it.

ItemStatusDetail
Australian owned and builtYesAustralian entity, developed in Australia
Runs entirely on your infrastructureYesNo vendor cloud in the request path
Air-gapped operationSupportedOffline licence file; no activation call required
Vulnerability disclosurePublishedNamed contact, acknowledged within two business days
Evidence you can checkPublishedResults on public datasets, with the harness, so your team can re-run them
ISM control responsibility matrixAvailable on requestFor your own system authorisation

Speed and safety

Measured, with the conditions named.

Figures without their test conditions are decoration. These are Shield's, measured on the CSIC 2010 corpus at engine level, and the boundary is stated with them.

~0.09 ms

added per request, verdict path

CSIC 2010 run, single core

91.69%

of attacking sources blocked by their third request

CSIC 2010

0

of 6,000 genuine sources wrongly blocked

same setting, same run

Fail open

if the process stops, traffic passes

default; fail-closed is opt-in

What the figures describe

These are per-source results: Shield watches a source's own requests until they agree, which is the shape credential stuffing, scraping and API abuse take. Source identity is configurable in the connector, so a fingerprint can be the subject rather than an address where your traffic calls for it.