Glossary

Short, canonical definitions for every term that appears across the product and docs. Each entry links to the full guide so you can dig deeper when you need to.

​

A

​

Action

A proposed or applied change to your infrastructure, such as a configuration update, a restart, or a pull request. Actions always carry an expected blast radius and follow the approval model set by Safety and Guardrails. See Action Center.

​

Action Center

The page where recommended actions are reviewed, approved, applied, and verified. See Action Center.

​

Agent

A specialised AI component inside the Agent Mesh. Each agent owns one slice of the reliability loop (observing, investigating, remembering, remediating, or enforcing policy).

​

Agent Mesh

The coordinated set of specialised agents that run the OPEL Loop together. See Agent Mesh.

​

Agent Skill

See Skill.

​

Agent Skills Store

The catalogue of system skills built by RubixKube plus the custom skills published inside your workspace. See Agent Skills Store.

​

C

​

Chat

See Rubix Chat.

​

Confidence score

A value attached to an insight, RCA link, or recommended action that reflects how much supporting evidence the system has. Higher scores mean less ambiguity and fewer plausible alternatives.

​

E

​

Environment

A connected unit of infrastructure. In the current release this is typically a Kubernetes cluster. Environments are the scope for Observers, signals, insights, and guardrails. See Environments.

​

Evidence

A concrete signal (log excerpt, metric sample, event, state diff) that backs a claim made by an agent. Every RCA link in a RubixKube report cites its evidence.

​

G

​

Guardian Agent

The agent responsible for policy enforcement. Nothing that mutates your environment reaches production without passing Guardian checks. See Safety and Guardrails.

​

Guardrails

The policies and safeguards that bound what agents are allowed to do, who needs to approve what, and how changes land in your environment. See Safety and Guardrails.

​

I

​

Insight

An anomaly or deviation that is worth a human’s attention. Insights are the top-level unit of work in the Magic Insights feed. Not every insight becomes an RCA. See Magic Insights.

​

Integration

A connection between RubixKube and an external tool (Slack, GitHub, PagerDuty, Notion, Linear, Atlassian Rovo, Teams, or your own services via MCP or REST). See Integrations.

​

K

​

Knowledge Graph

The live, queryable model of your infrastructure. Nodes are resources (services, deployments, pods, queues), edges are relationships (runs on, calls, depends on). The Observer Agent keeps it current. See Knowledge Graph.

​

M

​

MCP (Model Context Protocol)

An open protocol for exposing tools, resources, and capabilities to AI agents. RubixKube uses MCP for its internal agent-to-tool wiring and for custom integrations.

​

Memory Agent

The agent that stores incident history, resolution patterns, and operator context, then surfaces the relevant prior art when similar situations recur. See Memory Engine.

​

Memory Engine

The underlying store and retrieval system powering the Memory Agent. See Memory Engine.

​

O

​

Observer Agent

The only RubixKube component that runs inside your environment. Read-only, outbound-only, and deliberately lightweight. It streams signals to RubixKube Cloud. See Observer Agent.

​

OPEL Loop

The operational rhythm every agent in the mesh follows: Observe, Plan, Execute, Learn. See OPEL Loop.

​

R

​

RCA (Root Cause Analysis)

The process of tracing an incident from symptom to cause using evidence-linked reasoning. See Root Cause Analysis.

​

RCA Pipeline Agent

The agent that runs the RCA process on qualifying insights and produces RCA Reports. See Agent Mesh.

​

RCA Report

The evidence-linked output of a completed investigation: observed conditions, causal chain, recommended actions, and verification. See RCA Reports.

​

Remediation Agent

The agent that applies approved fixes inside their scoped blast radius and watches for the stabilisation window to close. See Agent Mesh.

​

Rubix Chat

The conversational front door to RubixKube, powered by the SRI Agent. Available in the console and through the Rubix CLI. See Rubix Chat Agent.

​

Rubix CLI

The terminal client for RubixKube. Install with npm install -g @rubixkube/rubix, launch with rubix. See the Rubix CLI docs.

​

RubixKube Cloud

The hosted control plane. Every agent except the Observer runs here, so your environment footprint stays small and upgrades are handled for you.

​

S

​

Signal

A single piece of infrastructure telemetry: a metric sample, an event, a log excerpt, or a state change. Signals are the raw material that feeds the Knowledge Graph, insights, and RCAs.

​

Skill

A reusable runbook the SRI Agent can run. Skills use the open Agent Skills format and can be enabled from the Agent Skills Store or published by your workspace. See Agent Skills.

​

SRI (Site Reliability Intelligence)

The product category RubixKube defines: AI-native reliability work that combines observation, investigation, and remediation into a continuous loop. See What is SRI?.

​

SRI Agent

The conversational agent that powers Rubix Chat. Answers questions, runs skills, and drafts actions, always with cited evidence.

​

System skill

A skill built and maintained by RubixKube, available to every workspace. System skills are read-only: clone one to adapt it for your team.

​

T

​

Tenant

A logical isolation boundary between customers. Each tenant has its own data, users, integrations, skills, and audit trail.

​

W

​

Workspace

The top-level container a tenant operates inside. Environments, users, integrations, and skills all live within a workspace.

​

Related guides