Deployment Options¶

The same platform runs on your infrastructure, not only on someone else's cloud.

SaaS¶

Cloud-hosted and multi-tenant. The fastest path to a first run, and where the free tier lives.

On-premises and self-hosted¶

Runs on your own cluster. Four components deploy together: a manager that reconciles the custom resources, a gateway that serves the API, a sidecar that performs secret injection, and an observer that collects telemetry.

graph TD
    G["Gateway<br/>HTTP API"] --> M["Manager<br/>reconciles the custom resources"]
    M --> SB["Agent sandboxes<br/>default-deny network policy"]
    SB --> SI["Sinjector<br/>secret injection sidecar"]
    SI --> VAULT[("Secret store")]
    SB --> OB["Observer<br/>collects telemetry"]
    OB --> G

Air-gapped¶

Runs with no internet connectivity at all. Logs, traces and audit evidence stay on your infrastructure, and AgentZ makes no outbound calls.

Bring your own LLM¶

Model-agnostic. Run on a private key or an existing subscription, against OpenAI, Anthropic, Google, Grok or a self-hosted open-weight endpoint. No markup and no proxy in the path. Connections and policy live on AgentZ rather than on the agent, so switching model does not mean re-wiring credentials or scopes.

Plans and compute limits at a glance¶

Free is the evaluation tier. Enterprise is priced on what you actually run. Full comparison in Plans and Compute Limits.