Per-task isolated sandbox for AI agents

A lightweight VM with its own kernel per agent task, reusing OCI images

Problem

More agents run code, and every task needs an isolated environment—but Docker's daemon, image bloat and memory pressure add up; one comparison found Docker shared storage at 363MB while lighter options cut it sharply. Developers want isolation without a daemon, on-demand, with low overhead.

Solution

A CLI tool that spins up a lightweight VM with its own Linux kernel per agent task from one command, reusing OCI images and familiar commands (run/build/exec) without a Docker daemon. Compression pool plus on-demand image reads keeps disk and memory low.

Users

Developers running multi-task AI agents and teams doing agent evaluation or automation

MVP

One-command lightweight VM with independent kernel; OCI image run/exec support; no daemon required; image compression and on-demand reads; basic resource comparison report

Revenue

Open-core plus enterprise features (team image pool, audit logs, multi-tenant quotas), or subscription by concurrent instances

Why now

Agent code execution is exploding, but Docker is heavy and cloud sandboxes are pricey; an indie can ship a usable CLI on top of lightweight virtualization in about two weeks