ORVYN
Download Get Started

Security

Every action ORION takes
is on the record.

An AI agent that can reach servers has to be accountable. ORVYN logs every connection, command, approval and file change, keeps secrets out of the model, and isolates each tenant and worker.

Download ORVYN Start Free

Defense in layers

Five boundaries between an agent and a mistake.

  • Autonomy levels

    Ask, Auto Read, Auto Workspace or Full Access — set per workspace or mission.

  • Approvals

    Production restarts, deploys and secret use can require a human, whatever the level.

  • Secrets by reference

    ORION works with SecretRefs; plaintext is decrypted only inside the runtime.

  • Isolation

    Tenants have their own boundary and keyring; each cloud worker runs one mission.

  • Audit trail

    Every event is timestamped with actor, resource and result.

Approvals

The agent stops. You decide.

When a step crosses a line you drew, ORION pauses and shows exactly what it wants to run, what it will affect and how it will be recorded.

  • The exact command

    You see the command and the target machine, not a paraphrase.

  • The expected impact

    Downtime, services affected and whether the action is reversible.

  • A scoped answer

    Approve once, approve for this mission only, or deny with a reason ORION will work around.

Permission matrix

Who can do what, at a glance.

Default roles in an ORVYN organization. Admins can tighten these with policies.

CapabilityOwnerAdminEngineerViewer
Start and run missions
Approve production actions
Connect servers and databases
Create and rotate secrets
Change agent policies
Read the audit log

Included Limited or on request Not included

Secrets Vault

The model never receives plaintext.

Credentials are stored encrypted in your tenant’s vault. ORION refers to them by name; only the isolated runtime that opens the connection decrypts them.

  • SecretRefs in prompts, plans and logs — never values
  • Per-tenant keyring and encryption at rest
  • Use is scoped to the mission and the resource
  • Every read is an audit event; out-of-scope reads are denied

What the model sees

{ "resource": "staging-web-01",  "credential": "secretref:vault/staging/ssh",  "action": "ssh.exec" }

Data handling

What stays where.

A plain description of how ORVYN treats what you connect.

Code and files
Read and changed inside your local workspace or the mission’s isolated cloud worker.
Resources
Servers, databases and cloud accounts are scoped per workspace and gated by role and autonomy level.
Actions
Commands, connections, approvals and file changes are written to the audit log.
Artifacts
Generated files stay in the workspace until you download or share them.

This page does not claim third-party compliance certifications. Enterprise customers can request current security documentation from KERNEL AI Labs.

Questions

Security questions

Is it safe to let an AI agent access my servers?

ORVYN is designed so it can be: autonomy levels limit what ORION may do, approvals stop risky steps for a human decision, credentials stay in the Secrets Vault, and every action is written to an audit log.

Does the AI model see my passwords or keys?

No. ORION works with secret references. Plaintext is decrypted only inside the isolated runtime that opens the connection and is never sent to a model.

What does the ORVYN audit log record?

Session opens, commands, approval requests and grants, service actions, secret reads (including denied ones) and file writes, each with time, actor, resource and result.

Can I require approval before ORION deploys or restarts services?

Yes. Organization policies can require approval for production restarts, deploys and other actions regardless of a mission’s autonomy level.

Is ORVYN SOC 2 or ISO certified?

This site does not claim any compliance certification. Contact KERNEL AI Labs for current security documentation.

Stop prompting.
Start delegating.

Give ORVYN the goal and stay focused on what matters.

Download ORVYN Start Free