The knowledge graph

At Joe’s center is a graph of your infrastructure. Its anchor nodes are the components you register — the external systems Joe manages. Around each promoted component, Joe’s background refresh discovers the resources that system exposes and adds them as nodes in their own right: for a Kubernetes cluster, for example, the workloads, services, and namespaces inside it; for a cloud account, the networks and instances under it. The edges connect these nodes into a model of how your environment fits together. The graph is persisted locally (SQLite-backed), and it is what gives Joe a model of your environment to reason over rather than treating every question as a blank slate.

Anchors and discovered resources

A component on its own is just a registered system — an anchor with nothing hanging off it yet. As Joe refreshes an armed component, it reads what that system currently exposes and records those resources as nodes beneath the component, reconciling them on each pass so the graph tracks reality rather than drifting. A discovered resource is a node in its own right, distinct from the component that anchors it.

The value is in the edges: this service runs in that cluster; this cluster’s metrics live in that Prometheus; this component’s alerts are handled by that alertmanager. Joe builds these relationships as it discovers and refreshes your infrastructure (see The agent loop and autonomy levels for how the background loop keeps the graph current), and they are what let Joe answer a question about one system by reaching the systems related to it.

Observability backends are resolved through edges

The most concrete payoff of the graph is how Joe answers observability questions. When you ask about a service’s metrics, logs, traces, or alerts, Joe does not require you to name the backend. It follows the graph edges from the subject of your question to the component that actually holds that signal — the metrics backend, the logs backend, the tracing backend — and queries there.

So “what do the metrics for this service look like?” becomes, internally, “which component is this service’s metrics backend, per the graph?” followed by a query against that backend. The edge is the resolution mechanism. This is why registering your systems and letting their relationships form is what makes Joe’s observability answers work without per-question backend wiring.

For connecting the systems that become components and their backends, see Components; for working with the graph through Joe, see Guides.