Observe only
The decision core is verified.
Audit mode returns Allow for a Baseline disposition, keeps logged true, and exposes the Allow enforcement preview.
Endpoint protection
See the decision before enforcement.
Audit mode evaluates a baseline executable locally, preserves execution, and exposes the tested enforcement preview.
Let the agent work. Keep the desktop reviewable.
Aran Cage creates a copy-on-write project twin, confines a trusted-but-fallible agent to declared zones, then gives you the exact divergence set to merge or discard. It is not a claim of an adversarial VM boundary or live two-way host sync.
Aran Cage local workflow
Setup neededA local Cage run keeps the host baseline read-only, records the agent's divergence, and requires review before changes return to the real project.
Step 1: Create a local project twin
The host baseline is read-only; unchanged content is shared until the agent writes.
Step 2: Run the agent in declared zones
The agent writes inside the Cage's tracked workspace rather than directly into the real project.
Step 3: Review, merge, or discard
Inspect the divergence set, merge only approved files, or discard the twin with the real project untouched.
One decision, exactly traced.
This sequence uses the shipped classifier and policy-decision tuple. It is not live enforcement or a persisted endpoint audit record.
Aran application-control decision
Setup neededA verified Audit-mode decision evaluates a baseline executable, then returns Allow with its enforcement preview.
Step 1: Audit-mode boundary
The verified decision uses Audit mode. Endpoint enrollment and effective-policy setup remain required; no public command is shown.
Step 2: Evaluate the baseline fixture
The tested C:/app.exe fixture resolves to the Baseline disposition through the local classifier.
Step 3: Inspect the decision fields
The decision returns Allow, logged true, and an Allow enforcement preview with reason audit: allowed + logged.
Know what is active.
Aran separates verified decision logic from endpoint setup, beta driver paths, and unavailable execution enforcement.
Setup needed
Activation requires administrator setup.
Use an enrolled Windows endpoint and effective application-control policy. No public Aran installer is shown.
Beta
Driver-backed protection stays platform specific.
Windows tamper protection and continuous data protection remain beta and are outside this decision proof.
Unavailable
Execution enforcement is not claimed.
The reviewed application-control path does not establish live process blocking, so this page does not claim an enforcement result.
Product boundaries
Aran Cage
Cage modules have separate platform gates. This observe-only decision does not claim Cage activation.
Keybound
Keybound is an Aran endpoint-security module. This workflow neither activates it nor claims a Keybound result, and Keybound is not a sixth public product.
The decision trace is product behavior evidence, not a certification, authorization, or compliance claim.
Delete the key line. The theft stops mattering.
Keybound seals your provider keys to the device's hardware. The application reads no key, so a leaked .env, a copied dotfile, or an exfiltrated disk image carries nothing anyone can replay.
From install to inert secret
Shipping for Java, Python, Node and .NETThe same three steps in every backend language: install the agent, delete the key line, and a stolen file stops being worth stealing.
Step 1: Install the agent
A small local daemon, installed once per host. It holds the hardware-bound identity; your application does not change yet.
Step 2: Delete the key line
Remove the line that reads the key from the environment. Add a process-entry hook — no source change — or one import if you prefer it explicit.
Step 3: The theft stops mattering
A leaked dotfile, a copied disk image, an exfiltrated .env — none of them carry anything a thief can replay, because the key never left the device.