Agents

Agents are configurable AI workers that run against your cluster. Each agent has its own instruction, scope, triggers, and tools, and files its findings as reports with actionable proposals that you review and approve. mogenius ships a set of curated default agents for common use cases — such as workload troubleshooting — and you can adapt them or create your own.

​

Where to find agents

Open the Agents section at either level:

• Cluster — agents and reports across every namespace the operator manages.
• Workspace — the same view, scoped to that workspace’s namespaces.

The landing page has two views, switched with the toggle at the top:

• Reports — the inbox of findings your agents have produced, with Open and All tabs.
• Agents — one box per configured agent, with its trigger summary, open‑proposal count, and last run.

​

Managing agents

From the Agents view, each agent box offers:

• Run now — Trigger the agent immediately, independent of its schedule or event triggers.
• Enable / disable — Turn the agent’s automatic triggers on or off without deleting it.
• Configure — Open the full configuration page.
• Selecting the box opens the agent detail page with its run history.

To create a new agent, choose New agent and fill in the configuration below. To remove one, open its configuration page and use Delete agent in the danger zone at the bottom — this is only available for existing agents and asks for confirmation before deleting.

​

Configuring an agent

The configuration page is organized into sections. A live preview shows the resulting Agent resource (mogenius.com/v1alpha1) as YAML.

​

Identity

• Display name — The human‑readable name shown in the UI.
• Resource name — The technical name (lowercase letters, digits, and dashes). Generated from the display name and fixed once the agent exists.
• Description — A one‑line summary of what the agent looks after.
• Icon — An icon for the agent box.
• Instruction — Free‑text guidance appended to the base system prompt on every run. This is where you describe what the agent should focus on.
• Enabled — Whether the agent’s automatic triggers are active.

​

Model

• Use cluster default — The agent uses the model marked as the cluster default.
• Specific model — Turn the default off to pick a particular model for this agent. Models are managed in AI Models & Setup.

​

Scope

Scope defines what the agent may access. Choose one of:

• Namespaces — Select specific namespaces, or enable All namespaces (*).
• Workspace — Reference a workspace; all of its namespaces are in scope.

​

Triggers

Triggers decide when the agent runs. You can combine them, and you can always run an agent manually.

• Cron — Periodic runs on a schedule. Presets cover hourly, daily, weekly, and monthly, or you can enter a custom 5‑field cron expression.
• On change — Runs when resources in scope change. Configure which kinds wake the agent (e.g. Pod, Deployment — leave empty for all watched types), which change types apply (created, updated, deleted), and a minimum interval (cooldown) to debounce noisy resources.
• Manual — Every agent can be triggered on demand with Run now in the UI, or from the CLI with mocli agent run <name>.

​

Tools

Tools determine what the agent can do. See Tools & Custom MCP Servers for details.

• Built‑in — Kubernetes — Read and manage Kubernetes resources in scope.
• Built‑in — Helm — Install, upgrade, and inspect Helm releases in scope.
• Custom MCP servers — Reference one or more custom MCP servers to give the agent access to their tools.

​

Budgets

Budgets override the model defaults for this agent:

• Max tool calls per run — Caps how many tool calls a single run may make.
• Token budget per run — Caps tokens per run (0 = unlimited).

​

Reports and proposals

When an agent runs, it produces reports collected in the Reports inbox. The inbox has two tabs:

• Open — Reports that still need attention (pending, in progress, proposed, or executing).
• All — Every report, including resolved and dismissed ones.

Reports carry a status that reflects their progress:

• Pending — Queued for processing.
• In progress — The agent is analyzing.
• Proposal — The agent has a proposed change awaiting your decision.
• Executing / Executed — An approved proposal is being applied, or has been applied.
• Solved — The proposed solution was accepted and the issue resolved.
• Ignored / Rejected — Dismissed by a user; hidden from dashboards and resource views.
• Failed — The report or its execution could not complete.

​

Approving proposals

When an agent is confident in a fix, it attaches a proposal — a create, update, or delete operation on a resource. You can:

• Show diff — Review a visual diff between the current manifest and the agent’s proposed version.
• Approve — Apply the proposal. mogenius executes the change and marks the report solved. Destructive operations (such as deleting a resource) ask for explicit confirmation first.
• Reject / Ignore — Dismiss the proposal without applying it.
• Solve — Mark the report as resolved manually.
• Replay — Re‑queue the report for a fresh analysis.

​

Agent runs and the reasoning timeline

The agent detail page groups findings into runs. Each run shows how it was triggered (event, cron, or manual), its duration, and token usage. Open a run to see the agent’s step‑by‑step reasoning timeline — the think → act → observe loop, including the tools it called and the findings it produced — updated live while the run is in progress.

​

Reports on a resource

When you view a specific Kubernetes resource, a banner appears if an agent has detected an issue with it. The resource’s report view is filtered to that resource, so you can review and act on findings in context.