HEAD

SYS://microVM  :: built for agents, not humans

The execution layer for your agents.

Every run is a secure microVM shaped to how agents actually work: fork any step into a parallel run, replay a failure exactly as it happened, suspend with live TCP connections held open, and hibernate for days, then resume in under a second.

Product Video

Roadmap

Shipped, hardening, and next.

States, not dates. What's in Now is live in the private alpha today. What's ahead is framed by the problem it solves, not a calendar.

Now4
  • Sandbox commit state machine

    checkpoint() writes full memory + disk into a browsable commit graph; fork() branches any commit into a parallel sandbox.

  • TCP-preserving suspend

    suspend() pauses a run with live TCP connections intact and resumes it in under a second.

  • Agent action observability

    Replay a run's filesystem diffs, process tree, syscalls, and network calls, step by step.

  • Multi-agent infra primitives

    Sandbox groups with shared workspaces and scoped credentials — the base primitives agent fleets coordinate through.

Building2
  • Multi-day hibernate

    Cold-tier pause for days with lifecycle hooks to re-auth on wake — hardening the wake path now.

  • Language SDKs from one contract

    Elixir SDK first; Python and TypeScript generated from a single OpenAPI / gRPC contract, CLI alongside.

Next2
  • Deeper multi-agent coordination

    Shared workspaces and scoped delegation that make fleets of agents safe to run together.

  • Richer run observability

    The visibility layer between model traces and infra metrics, so you debug an agent like you debug code.

Status

Service Health

Status here comes from live monitoring against a real endpoint — never a hand-set dot. The control plane and the sandbox runtime are reported separately, because they are at different stages.

Monitoring coming online

Control plane / API

Elixir/BEAM-backed. Scheduling, state DAG, and the API surface.

Real monitoring being stood up.

Monitoring coming online

Sandbox runtime

MicroVM execution. Private alpha, hardening on reserved hardware.

Real monitoring being stood up.

Frequently Answered Questions

What engineers ask us first.

The questions that come up in the first ten minutes of every call, answered the same way here.

What exactly is a sandbox here?

A secure microVM built for a single agent run — its full memory and disk versioned as a commit graph. Every step is a commit you can fork, replay, suspend, or hibernate: the four moves above.

How does suspend keep TCP connections alive?

The sandbox is paused with its live TCP connections held open, not torn down — sockets are part of the state that gets preserved, so on resume they are still there. Bring a connection-heavy workload to a call and we'll demonstrate it live.

What's underneath — containers or VMs?

MicroVMs. Each run gets VM-level isolation, not a shared-kernel container. We walk through the runtime stack, isolation boundaries, and threat model on a call.

Where does it run? Can I self-host?

The private alpha runs on reserved hardware we operate, with white-glove onboarding. No hosting vendor is locked yet, so deployment options are genuinely open — if self-hosting matters to you, raise it early and it shapes the roadmap.

What does it cost?

Indicative alpha rates are on the pricing page — directional, not a quote. The model: pay for active compute, while idle sandboxes hibernate on a cold tier that costs almost nothing. Real numbers come out of a call.

How does MRIN compare to X?

The sandbox-runtime space has been growing fast. MRIN's contribution is on several primitives stacked on a full Linux microVM operator surface (REST + auth + TLS + audit + metrics). we have a docs section sketches how MRIN compares to the projects most worth benchmarking against. We have an honest comparison to existing solutions in the space.

What's actually live today?

The roadmap is states, not dates: everything under Now runs in the private alpha today, Building is being hardened, Next is direction. If a capability matters to you, ask for it on the call and we'll demo it — or tell you honestly that it isn't there yet.

Get Started with Mrin.

Private alpha, white-glove onboarding. Bring a workload and we'll fork, replay, and suspend it live on a call.