StackMap
Subscribe

forkd vs tensorlake

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. — versus — Serverless platform for agent sandboxes: stateful Firecracker microVMs with snapshots, cloning, auto suspend/resume and network policy, plus fan-out orchestration functions. Python SDK and CLI.

The curated verdict

Both make sandbox state forkable: forkd forks copy-on-write children from a warm Firecracker parent on your own KVM hosts, Tensorlake snapshots and clones running sandboxes in its cloud.

forkdtensorlake
Stars2.9k1.0k
Forks226148
LanguageRustPython
LicenseApache-2.0Apache-2.0
Last activity14 days agotoday
Topicssandboxes, agents, localsandboxes, agents
Curated connections54

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.

tensorlake — the curator's take

Pick Tensorlake when you want hosted sandboxes that behave like real machines: stateful Firecracker VMs that suspend when idle and resume with memory intact, snapshot and clone mid-run, live-migrate, and take per-sandbox egress allowlists, plus a serverless function runtime to fan out agent work with each function in its own sandbox. This repo is the SDK and CLI; the runtime is Tensorlake Cloud, so it is an API key, not something you self-host. The filesystem benchmark against E2B, Modal and Daytona is their own. Need it on your own hardware? cubesandbox or agent-sandbox. Running one agent's code locally? A container may be enough.