Getting started#
Mirantis Agent Platform is one Helm chart: the control plane, the sandboxes it runs, the database, and the audit trail. You can install it yourself three ways: the Quickstart on one VM; the production install on your own Kubernetes cluster; or on your laptop.
There is no sales call to book first, no environment to request, and no evaluation key to wait for. A new install runs a full 30-day evaluation as soon as you activate it, and you can sign in with a Lens ID you create yourself in a minute, or with your own identity provider.
Start here#
-
You bring an empty VM and an LLM key, and the guide sets up the cluster, address, and certificates for you.
-
On your own Kubernetes cluster, with your own ingress, database, and sign-in.
Other ways to try#
- On your laptop (step by step, or with a coding agent) — runs on your own machine and needs local Kubernetes tooling (
docker,minikube,kubectl,helm).
What you need#
Every path needs a way to sign in — a Lens ID account, which the Quickstart and the laptop install use (sign up at app.k8slens.dev), or your own OIDC identity provider — and an inference-provider credential for managed inference. An OpenRouter API key is the shortest to obtain; Supported providers lists every backend. The rest depends on the path:
- Quickstart (VM) — an empty Ubuntu VM with a static public IPv4 address, and a cloud firewall you can open ports in. See its prerequisites.
- Install for production — a Kubernetes cluster you can install into, and Helm. See its prerequisites.
- On your laptop —
docker,minikube,kubectl, andhelm. See its prerequisites.
What you end up with#
A running platform. The Quickstart and the laptop install go on to launch Prism Agent, Mirantis Agent Platform's ready-made managed agent, which answers you in a browser tab from inside a sandbox that has the network denied by default, holds no real credentials, and writes every action it takes to the audit trail. The production install ends with the install activated, ready for your first agent.
From there, the governance chain is yours to drive: attach a policy, connect the systems the agent must work with, and watch what it may and may not reach.
Related#
- Core concepts — the architecture, the sandbox isolation model, and the identity model
- Sandboxes — creating, running, and exposing agents
- Administration — org policy ceilings, bindings, project members, tokens, spending limits, and audit
- Security — threat model, compliance posture, and the security whitepaper
If you want a hand#
Self-serve is the main path, and it stays that way. But some things are worth a conversation: an air-gapped install, microVM isolation with a Kata Containers or gVisor runtime class, SSO against your directory, or a policy set that has to satisfy a specific control framework.
Talk to us and we will scope it with you — architecture review, security-model walkthrough, and a production-shaped environment configured alongside your team. Install first if you prefer; the conversation is more useful when you already know what you are asking about.
See also Support for the full set of contact channels.