FAQ¶

Real questions from teams evaluating AgentZ.

Does AgentZ run in our cloud or yours?¶

Standard tiers run on shared cloud. Enterprise gets VPC, on-prem or air-gapped deployment with a private tenant. Either way the sandbox policy engine runs local to the agents. It is not a remote proxy making decisions over the wire.

What stops an agent from calling an API it should not?¶

The sandbox enforces a default-deny network policy. Every outbound call is checked against an explicit allowlist before it leaves, and anything blocked lands in the audit trace with the domain and port it tried to reach.

Will it work with the model and framework we already use?¶

Yes. AgentZ accepts any model provider on your own key: OpenAI, Anthropic, Google, or a self-hosted endpoint. Wrap an existing LangGraph or CrewAI agent, or build skills natively. The governance layer sits underneath either way.

How does the audit trail hold up in a compliance review?¶

Every tool call, memory read and model response is recorded and stored with a deterministic replay id, so a run can be replayed exactly as it happened. Retention varies by plan and Enterprise is custom. TODO: confirm the exact windows.

What is on the free tier and what costs extra?¶

Free covers a small compute quota, unlimited skills and sandbox guardrails. Prompt firewalling, RBAC and teams, longer retention and dedicated compute need a paid plan. TODO: reconcile the seat count, which the pricing table gives as 2 and the public FAQ as 3.