Question 1

Do you store our agent credentials?

Accepted Answer

No. Kernel ships as an SDK that runs in-process inside your backend. Your secrets, API keys, OAuth tokens, and customer data never leave your infrastructure. We see policy decisions, not the data the agent is acting on.

Question 2

What's the latency overhead?

Accepted Answer

Under 10ms p99 for the policy check. Decisions are deterministic, not prompt-based. Kernel does not ask an LLM whether an LLM should be allowed to act.

Question 3

How long does integration take?

Accepted Answer

Most teams ship the first integration in a single sprint. The check() call itself is three lines. The real work is mapping your highest-risk actions and writing the policy, and we walk you through that in the first week of onboarding.

Question 4

How does Kernel fit with tools we already use?

Accepted Answer

Identity tools answer 'who is this agent?' Observability tools answer 'what did it do?' Kernel sits between them and answers 'should this specific action execute right now', in real time, before it happens.

Question 5

Why not write our own governance logic in-house?

Accepted Answer

You can, and most teams start there. Consistency across a growing fleet, EU AI Act Article 12 audit requirements, edge cases, and AIUC-1's third-party governance requirement make in-house hard to maintain and, for some companies, non-compliant.

Question 6

Why not use an LLM to govern our agents?

Accepted Answer

Using a probabilistic system to govern a probabilistic system produces non-deterministic decisions, can itself be prompt-injected, and won't satisfy regulators. Governance decisions must be deterministic and explainable.

Question 7

Can we write our own policies?

Accepted Answer

Yes, three ways: plug-and-go templates, co-development with our team for vertical-specific logic, or a conversational AI Policy Agent that walks you through writing and validating a custom policy.

Question 8

What does the audit log give us for compliance?

Accepted Answer

Every decision (ALLOW, DENY, REQUIRE_HUMAN) logged immutably with full chain context. Queryable, exportable in EU AI Act format, with plain-language decision narratives. Covers EU AI Act Articles 9, 12, 13, 14, DORA ICT audit requirements, and AIUC-1.

Question 9

What happens after the demo?

Accepted Answer

If we're a fit, we get you into a paid pilot, typically 4 to 6 weeks, one workflow, a draft policy your team co-owns by the end.

Ship the agents you've been holding back.

Frequently asked questions