Launching today

npm i -g hotcell
Local sandboxes for AI agents on your Mac, Linux, bare metal
82 followers
Local sandboxes for AI agents on your Mac, Linux, bare metal
82 followers
Self-hostable sandbox SDK inspired by Cloudflare Sandbox SDK. works on any device (Mac, Linux, bare metal). super easy to use (both by human engineers and agents). you get full control over capacity/token spent per sandbox, you can default-deny egress but still allow access to LLM providers, and API keys never directly enter each sandbox (it creates per-sandbox tokens instead that die when the sandbox stops). npm i -g hotcell






npm i -g hotcell
What's the isolation model - containers, lightweight VMs, or something custom? Curious how heavy it feels on a laptop running a coding agent alongside the sandbox.
npm i -g hotcell
@kritishpuri not custom. 3 drivers: containers (docker) by default, or microvms, firecracker on linux/kvm and apple VZ on macos. you can set a default once at setup and can override it per sandbox (--driver), so a single daemon can run both tiers at once.
re: weight, here are some numbers (all in docs/benchmarks.md):
idle footprint measured ~54 mb of host ram per firecracker vm (40 on a 4-vcpu box).
on a real workload (clone + install + typecheck) virtualisation cost was ~16%.
StepGrab
Sandboxing agents locally makes a lot of sense to me. How much does the isolation cost you in practice on a Mac? Curious whether it's noticeable or whether it disappears into the noise.
the per-sandbox scoped tokens that die with the sandbox is honestly such a clean move, keeps api keys out of the blast radius without making the workflow annoying