StackMap
Subscribe

CubeSandbox vs forkd

Hardware-isolated microVM sandboxes for AI agents — sub-60ms boot, <5MB overhead, E2B-compatible API, self-hosted on your own KVM nodes. — versus — fork() for agent microVMs: children fork copy-on-write from a warm Firecracker parent — 100 KVM-isolated VMs in ~100ms, live-VM branching in ~56ms, portable snapshots from a hub.

The curated verdict

Both are self-hosted Firecracker-class microVM sandboxes for agents. CubeSandbox sells a platform (E2B-compatible API, sub-60ms cold boots); forkd sells a primitive — copy-on-write fork from a warm parent, which wins precisely when 100 children share one expensive warm state.

CubeSandboxforkd
Stars11k2.7k
Forks946212
LanguageRustRust
LicenseNOASSERTIONApache-2.0
Last activityyesterday2 days ago
Topicsagents, localagents, local
Curated connections101

CubeSandbox — the curator's take

Reach for CubeSandbox the moment your agents execute model-generated code and "just run it in Docker" stops feeling safe — it gives every tool call a disposable hardware-isolated microVM with E2B's SDK ergonomics, minus the SaaS bill, plus snapshot/rollback of any sandbox state. The catch: it's real infrastructure — you need KVM-capable Linux hosts and someone willing to operate them. Prototyping a single local agent? A container or E2B's hosted tier is less machinery. It's a runtime, not a framework — you still bring LangGraph/AutoGen/whatever on top.

forkd — the curator's take

The sandbox primitive agent fan-out actually wants: instead of cold-booting a kernel per child, a parent VM boots once with your runtime warm (deps imported, model loaded, JVM JITed) and children mmap its memory copy-on-write — so spawning 100 isolated environments costs ~100ms, and v0.4 can BRANCH a live running VM. Tree-search agents, parallel experiment runners and RL rollouts are the natural fits. NOT a managed platform: Linux/KVM only, you operate it, and there's no E2B-style API compatibility — this is a runtime primitive you build on, not a sandbox service you call.