Platform

Aperture: one policy engine for everything

Write rules once; Aperture compiles them into enforcement across the mesh, egress, and AI endpoints — with every decision logged and every change versioned.

Precision iris metering light beams toward an AI processing core

Policies that read like sentences

Declare intent once — "finance may call the forecasting model, only from managed devices" — and Aperture enforces it on every path. No dialects per layer, no dashboard-only config that escapes review: policy lives in files, gets reviewed in pull requests, and deploys atomically.

# finance may call the forecasting model, only from managed devices
group:finance + device:managed -> model:forecast allow
 
# research clusters may reach their vector store and nothing else
group:research -> db:vector-prod allow
group:research -> tag:prod deny
 
# unsanctioned AI providers: log, don't block (yet)
group:eng -> ai:unsanctioned log

Enforcement on every path

The same rule set governs mesh traffic, egress routes, and AI gateway calls. A change applies immediately everywhere — no redeploy, no agent update, no waiting for the next release train — and every evaluation is attributable to a rule, a revision, and a person.

enforcement — coverage
meshnode-to-node trafficacl v182ENFORCING
egressoutbound allow-listsacl v182ENFORCING
ai gatewaymodel endpointsacl v182ENFORCING
dnsname resolutionacl v182ENFORCING
legacy vpnperimeter rules—RETIRED
One revision, every path — atomic deploys, instant rollback.

What you get out of the box

Governance primitives that compose — no separate product per layer.

Human-readable policy

Rules read like sentences. Reviewers argue about intent, not syntax.

Policy in git

Files under version control, reviewed in pull requests, deployed atomically.

Instant apply & rollback

One edit takes effect immediately; reverting is as fast as the edit.

Testable rulesets

Dry-run a change against last month's traffic before it ships.

Outbound allow-lists

Workloads reach exactly the hosts they need — registries, telemetry, APIs — and nothing else.

Regional exits

Pin egress to a compliant region for regulated data paths.

Provider isolation

Each vendor sits in its own scope; a leaked token never becomes lateral movement.

Rate & spend guards

Per-team ceilings catch runaway agents before the invoice does.

Rule-attributed decisions

Every allow, deny, and log names the rule and revision that made it.

Change provenance

Who edited which rule, when, and with whose review — automatic.

SIEM streaming

Ship decisions to Datadog, Splunk, or any webhook endpoint.

Exportable history

Full decision and policy history as JSON or CSV, on demand.

Questions and answers

Does Aperture require the whole OpenVLAN stack?
It's strongest on the mesh — where identity is proven at connect time — but the AI gateway and egress controls run standalone against your existing infrastructure.
What happens if two rules conflict?
Deny wins, and the linter flags the conflict before you can ship it. Policy files fail closed in review, not in production.
Can non-engineers write policy?
The syntax is plain enough for security and finance teams to read and review, while engineers own the pull-request workflow. That split is the point: readable by everyone, mergeable by process.
How do we test a rule before it goes live?
Dry-run mode replays proposed rules against recorded traffic and reports what would have been allowed or denied — a preview, not a gamble.
Does enforcement add latency?
Decisions evaluate in microseconds at the connection point, not per packet in a middlebox. Data keeps flowing on direct paths at line speed.
Can we run it in monitoring-only mode first?
Yes — ship every rule as log before flipping to allow or deny. Watch what would happen for a week, then tighten with evidence.

One engine. Every path. Full audit.

Write your first three rules in the time it takes to read this page.