Part 0
ax vs kagent
Two Kubernetes agent platforms that both run on Agent Substrate but answer different questions. This page sets out who the tutorial is for, what you will learn, and the short answer.
Who this is for
This tutorial is for platform engineers who already work with Kubernetes custom resources and kubectl, and who need to choose between, or run alongside each other, two agent platforms built on Agent Substrate.
Reading the concept and comparison pages takes about 45 minutes. The two labs add roughly an hour each, including a kind cluster per lab: ax runs on upstream Substrate v0.3.0, and kagent alpha6 requires kagent's own Substrate fork, v0.3.0-alpha3.132
What you will learn
After this tutorial you can:
- Explain which parts of sandboxing, suspend, resume, and snapshots come from Agent Substrate rather than from ax or kagent.
- State who owns the agent loop in each project, and what that means for the code you ship.
- Deploy a first ax
Taskand a first kagentAgentwith aSession, and recognise success from the command output. - Pick a project for a given workload using the decision flow, with each answer traced to evidence.
The short answer
ax schedules sandboxed agent workloads, while kagent defines, hosts, and talks to agents.18
ax
ax is a declarative orchestrator for agent workloads that runs on Agent Substrate and is built to run "billions of tasks per cluster".1 You declare a Task with an image, a command, and limits, plus a Workspace with Git repositories, MCP (Model Context Protocol) servers, and skills.3 Your container brings its own agent loop, and ax keeps its state in Redis rather than in etcd.23
kagent
CNCF Sandbox
kagent is an agent platform. An AgentTemplate says what an agent does, a Harness says how it runs, and an Agent pairs one of each.25 kagent supplies the loop, conversation history in PostgreSQL, A2A (Agent2Agent protocol) and MCP endpoints, and a web UI.8 Each conversation is a Session that you can suspend, resume, checkpoint, and fork.810
Tour of the tutorial
This one-minute narrated recording shows the theme switch, sidebar navigation, a lab page, the diagrams, and the mobile contents drawer. A transcript follows the video.
Transcript
This tutorial compares ax, Google's Agent Executor, with kagent. Both run agents on Agent Substrate, but they answer different questions.
You can switch between light, dark, and system themes, and the sidebar always shows the active mode.
The start page sets out who the tutorial is for, what you will learn, and the short answer.
The sidebar walks you through each part. This is the ax lab, with manifests, commands, and the output you should expect.
Each lab includes diagrams. This one shows what the ax task runner does inside the sandbox before your command starts.
The architecture page puts both control planes side by side, with Redis behind ax, and PostgreSQL behind kagent.
On a phone, the contents move into a drawer, and wide diagrams switch to a vertical layout. The decide page ends with a question flow to help you choose.
How the tutorial is organised
The pages run from concepts, to hands-on labs, to side-by-side comparison, to a decision.
| Stage | Pages | What you get |
|---|---|---|
| Concepts | 1 Substrate, 2 Agent loop | The shared layer and the core split between the projects. |
| Do | 3 Setup, 4 ax lab, 5 kagent lab | One checklist, then a first workload in each project, each on its own Substrate build. |
| Compare | 6 Architecture, 7 Resources, 8 State, 9 Lifecycle, 10 Interfaces, 11 Maturity | Components, resource mapping, state placement, lifecycle, APIs, and release status. |
| Decide | 12 Recap, 13 Decide | The three biggest differences and a question flow. |
| Reference | 14 Glossary, 15 Sources | Term definitions and every cited source with its pinned commit. |