Starting point
Kubernetes state is often hard to see for teams, spread across kubectl output, logs and dashboards. ranching.farm collects live cluster data and has an LLM explain it instead of just displaying it.
ranching.farm — AI-assisted Kubernetes insight
ranching.farm is a platform for which Franz Bettag builds the Elixir/Phoenix/LiveView backend and the Kubernetes agent integration. A Go agent inside the cluster executes tools over an authenticated Phoenix channel; the platform checks role-based permissions and Aquila orchestrates the LLM with tool results.
Problem
Role
Franz Bettag's role
Builds the platform (Elixir/Phoenix/LiveView) and the Kubernetes agent integration.
Architecture
Backend
Elixir/Phoenix/LiveView platform with Aquila for LLM orchestration.
Cluster agent
A Go agent runs inside the cluster and connects over an authenticated Phoenix channel; cluster selection is scoped per organization.
Tools
The platform checks tool calls before forwarding them: ordinary team roles are limited to allowed read commands; admins and owners can also execute modifying commands. The Go agent executes delivered commands. The LLM interprets the results and streams a response.
Results and limitations
What's documented
A public demo and the open-source agent code are available. The agent itself is not a read-only sandbox: platform permissions and Kubernetes RBAC determine which changes are allowed. No percentage or customer-count metrics have been collected.
Not independently measured
Operational time savings are not independently measured or verified.
Evidence
Sources
Public demo and open-source agent code.