Sandboxes for AI via WASM, without server-side compute
Rewind a run of your agents' code to any moment, and step through it.
Stage
Established
Not given
Founded
2015
Not given
Based in
GB
Not given
Pricing model
Freemium
Not given
Pricing
Free plan: $0/month, 10,000 tokens/month (up to 1,000 hours of compute). Pro plan: $20/month with 20,000 tokens/month (up to 2,000 hours), then $0.01 per hour overage. Enterprise: Custom pricing. Token top-ups: $10 per 10,000 tokens on paid plans. Open-source: up to 100,000 tokens/month. Startups: u
Not given
Key features
Linux-compatible kernel running via WebAssembly on user devices
Support for Node.js, Rust, and Python (preview) runtimes
POSIX-like filesystem backed by OPFS and IndexedDB
Portals for exposing sandbox services via public URLs
Git, npm, and bash shell access inside sandboxes
Millisecond boot times with no provisioning required
Unlimited concurrent sandboxes per user
Code isolation in browser sandbox prevents access to local files
Persistent or ephemeral filesystem per instance
Supports AI coding agents without exposing infrastructure
Rewind to any instruction with thread stacks and memory inspection
Replay identical runs across different machines
Runs unmodified Docker images with custom kernel and hypervisor
1000 times more efficient logging than traditional tracing
Audit all external events with configurable security rules
What makes it different
Zero server-side infrastructure needed; compute runs entirely on user devices via WebAssembly
$0.01 per hour compute cost with 1,000 free hours monthly, significantly cheaper than cloud sandboxes
No resource contention or degradation under load since each session uses its own device
Built-in browser security boundary isolates untrusted code from host machine
Instant project sharing via Portals—demos are live, working builds not videos
Records and replays every instruction, memory value and packet instead of sampling events
Secure by design with kernel and hypervisor written together from scratch
Deterministic replay works identically across machines without rerunning code
Full visibility into execution including all thread stacks and memory addresses
dvm