Ultra13
Platform · Memory Safety

How does Ultra13 keep AI agent memory safe?

Ultra13 applies a write gate and a recall gate to agent memory. The write gate decides what is allowed to become durable memory and labels it by source and trust class; the recall gate decides what stored memory may influence. Untrusted memory can shape a preference but cannot authorize a tool call, export, or approval.

By the Ultra13 teamPublished Updated
TL;DR
  • Write gate: control what becomes durable memory, labelled by source and trust.
  • Recall gate: control what stored memory may influence.
  • Separate preferences from operational instructions.
  • Untrusted memory can never authorize a privileged action.

Why memory needs gates

Memory persists across sessions, so a single poisoned write can steer future users or tenants. Background: memory poisoning in AI agents.

FAQ

Frequently asked questions

Can Ultra13 quarantine risky memory writes?
Yes — the write gate can quarantine or require approval for writes derived from untrusted sources, and expire risky entries.

See where the firewall stops the path.

Give us one agent workflow. We’ll map the context boundary, replay the abuse paths, and show where the Context Firewall blocks them.

Join WaitlistSee a Sample Report