Sandboxes — inside the Agent Runtime

A box of its own for every run.in development

An agent that can touch files and run tools needs somewhere safe to do it. Sandboxes will give every run its own sealed working space — real tools inside, torn down clean when the work is done. It's a piece of the agent runtime we're building now, on top of the layer that's already live.

What a sandbox is[01/02]

An isolated place to do the work.

Today you hand an agent a box you own and hope it cleans up after itself. The sandbox will make that our job — a fresh, isolated environment for each run, with everything the agent needs inside and nothing of yours at risk.

Status: in development
SB-01

One box per run

Every agent will get a working space of its own — isolated from every other run, from your machine, and from ours. What one agent does inside its box will stay inside its box.

SB-02

Real tools inside

A shell, a filesystem, packages to install, room to write and read and try again. The sandbox will be somewhere the agent can actually do work — not a stubbed-out toy environment.

SB-03

Sealed and ephemeral

The box will open for the run and close when it's done — torn down clean, nothing left behind, nothing leaking out. No long-lived boxes to patch, no stray state between jobs.

SB-04

Nothing of yours to run

No VMs to provision, no containers to babysit, no cleanup cron. The sandbox will be part of the runtime — spun up and reclaimed on our side, on the key you already hold.

Where it sits[02/02]

Part of the runtime. On proven ground.

Sandboxes are one piece of the agent runtime — Layer 02 of the platform — which stands on the inference layer serving traffic in production today. When your agents get a box, it won't be floating on something new.

While it's built

Start on the inference layer today — your key will carry over.

Put your agent on real infrastructure

Your agent doesn't change.
Everything underneath does.

$export ANTHROPIC_BASE_URL=https://api.layerx1.in

Start free · no card · Starter from $5/mo