Core Concepts¶

The building blocks, and how each one relates to the others. Every term here is used with exactly this meaning throughout the documentation.

Agents, workflows and skills¶

A skill is a reusable, versioned building block. A workflow chains skills into steps. An agent is the compute that runs them.

Sandboxes¶

Isolated compute for one agent: its own vCPU allocation, memory and filesystem, wrapped in a default-deny network policy. In the self-hosted build this is a Kubernetes custom resource that the manager reconciles.

Credentials¶

Injected at runtime and never exposed to the model. On credential creation you store a placeholder that is resolved at request time, so the model never receives the real secret and cannot store it.

Workspaces¶

The isolation boundary for a team. On creation you pick a type, General or AI-SOC, and an AI-SOC workspace surfaces integrations, triggers and investigation views that a General one does not.

Triggers¶

Three ways to start a workflow: an inbound webhook, a cron schedule, or a manual run.