Meshery is core to the platform I engineer at work, and I have been enjoying it more with every month that AI agents become a normal part of how my team operates. The reason is simple: Meshery was built around a model of infrastructure (components, relationships, designs, connections) that an agent can reason about just as well as a person can. The tool that makes that obvious in daily use is mesheryctl-axi.
What "AXI" means here
AXI stands for agent experience interface. It is a thin wrapper around an existing CLI that reshapes the output so it is cheap and unambiguous for an AI agent to read, and still pleasant for a human. mesheryctl-axi wraps mesheryctl and gives it a small, discoverable surface.
$ mesheryctl-axi --help
usage: mesheryctl-axi [command] [args] [flags]
commands[5]:
(none)=home, connection, system, design, model, component
flags[2]:
--help, -v/-V/--version
requires:
mesheryctl installed and authenticatedFive command groups, and a home view when you give it nothing. That is the entire mental model, and it fits in the first screen of output.
Why this works so well for agents
- Compact output. Results are structured and terse, so an agent spends its context window on the task, not on parsing banners and tables.
- Self-describing. The home view and the examples in
--helptell an agent what to run next. There is no guessing at flags. - Predictable formats. When you need full content, you ask for it explicitly with
--format yamlor--format json. - Reads first. The commands I reach for most are read operations that are safe to let an agent run freely.
# What is the state of my Meshery server?
mesheryctl-axi system status
# What clusters and cloud accounts does Meshery know about?
mesheryctl-axi connection list
# What designs exist, and what is inside one?
mesheryctl-axi design list
mesheryctl-axi design content my-design --format yaml
# What can Meshery model, and what are the components?
mesheryctl-axi model content kubernetes --format json
mesheryctl-axi component listA workflow I actually use
When something looks off in a cluster, I now start by asking an agent a plain-language question and letting it drive mesheryctl-axi. It checks system status, lists the connection entries to find the right cluster, pulls the relevant design in YAML, and tells me where the running state and the intended design diverge.
That is not a demo. It is a few seconds of reading that used to take me a handful of terminal tabs and a lot of scrolling.
mesheryctl-axi update --check reports your version against the latest published one, and mesheryctl-axi update upgrades it. Set MESHERYCTL_BIN if your mesheryctl binary is not on your PATH.
Why I call Meshery AI-centric
Plenty of tools bolt a chat box onto an old interface. Meshery's strength is underneath. Everything is a typed, relationship-aware model: a component knows what it can connect to, a design is a document that describes a whole system, and a connection is a first-class thing. Those are exactly the shapes that agents are good at working with. A wrapper like mesheryctl-axi only has to get out of the way.
In the next post I will cover what that has meant for the teams who build on the platform.