Zeroport Agent
Sovereign AI on-device that watches everything that crosses the air gap. The moment it recognizes intent that breaks your rules, it ends the session physically.
It lives inside the appliance, on site.
Zeroport Agent runs inside the Fantom Edge or Enterprise yet out of band from the sessions it watches.
It runs inside the Fantom Edge and the Fantom Enterprise.
Everything is processed there. Session content never leaves your site.
Watch it judge.
An AI agent has finished its task. Then it starts to type something the rule forbids.
Zeroport Agent, shown as a black ball with a violet symbol, watches an AI agent type a sentence that breaks the rule. Before the sentence is finished the symbol turns red, the session ends and the words break apart.
It watches anything that crosses.
The screen, frame by frame. Every keystroke and every mouse inputs. It makes no difference whether the hands on the keyboard are a person's or an AI agent's.
It reads a session the way an analyst does.
Reads the screen.
A terminal, a browser, a document: it knows the difference. It reads the command on screen and sees the output that came back.
One timeline.
Keystrokes and clicks sit on one clock with the screen, so it knows whether a command was typed, run or already showing results.
Knows the journey.
It follows each session from the first login and keeps a running summary.
Rules are intents seen as actions, not phrase matches.
No customer data leaves through personal email.
You write a rule as one plain sentence. It becomes a staged detector that Zeroport Agent talks through with your analyst and tests against your past recordings before it is armed. Every session is then judged in context, so the same screen can be fine in one session and a violation in another.
The cut is physical.
Keystrokes and mouse movements go into your network as light through the Fantom Core. The moment Zeroport Agent recognizes intent that breaks a rule, that light is physically shut off. The session ends at once, and the prompt or the action in flight is gone with it.
Five steps. You set the dial.
- Observe
- Alert
- Investigate
- Terminate
- Self-tune
Triggers and escalation are fully configurable: Flag, alert, investigate, terminate, revoke certificate - you choose the threshold and the response per rule. Every action lands in the audit log before it runs, with full evidence trail retained and displayed.
AI agents are judged on four routes.
Your own AI agent works through Zeroport in your name, with an agent certificate bound to you. Zeroport Agent judges both of its conversations, both ways.
- Ask
- You to your AI agent
- Answer
- Your AI agent to you
- Type
- Your AI agent to the session
- Read
- The session to your AI agent
Each item is held until it is judged. If no verdict comes, it is blocked. The rules are the same natural language rules, with a verification and an action on each route. When you and your AI agent work simultaneously, each works via its own Fantom Core.
Nothing leaves.
Everything is processed on-device, at your network boundary. Its policy engine sits out of band, where no session can reach it. Both models carry it.
Specifications
- Runs
- Inside the Fantom Edge and the Fantom Enterprise, on your site
- Sees
- Every keystroke and mouse move going in, every frame of screen coming back
- Judges
- Each session in context, against rules written as plain sentences
- People and AI agents
- Treated the same way
- AI agent routes
- Ask, answer, type and read: held until judged, blocked without a verdict
- Autonomy
- Five steps: observe, alert, investigate, terminate, self-tune
- Ends a session
- Physically: the light inside the Fantom Core stops passing
- Rules
- Plain language, tested against your history before they go live
- Record
- Every action in the audit log, with the evidence
- Leaves your site
- Nothing
Tell us about your network and the agentic work challenges you face.
Next product: Fantom Edge
25 Fantom Cores in scaled-down 1U, for smaller sites and branch networks.