USERLITE / AGENT-NATIVE EXECUTION
The runtime for code no human wrote.
Agents generate and execute code constantly. rayJS runs it with near-zero startup, a footprint measured in kilobytes, deterministic results, and zero access to anything by default.
002 / INTERNAL MEASUREMENTS
Near-zero is the new baseline.
Agent systems execute code constantly. rayJS removes the cold-start and memory tax that turns every small decision into infrastructure.
From launch to execution in less time than most systems take to notice a request.
A radically smaller footprint than general-purpose JavaScript infrastructure.
Tens of thousands of isolated executions launched in seconds.
Closing in on V8.
rayJS now runs our internal suite within roughly 1.7× of V8 — while keeping the startup, footprint, determinism, and isolation agent platforms need.
Figures shown are directional internal measurements. Workload, environment, and warmup affect results.
003 / LOCKED DOWN BY DEFAULT
Nothing gets in. Until you let it.
Every execution starts with zero access. Add only the capabilities the workload needs — an input, an approved API, a specific operation — and nothing else.
fs.readFile("/etc/passwd")DENIEDnet.connect("0.0.0.0:443")DENIEDprocess.envDENIEDchild_process.spawn("sh")DENIEDfetch("api.approved.example")GRANTED
Zero ambient authority
No filesystem, network, database, or system access unless you provide it.
Exact capability control
Expose the smallest useful surface for each task and each agent — an input, an approved API, a specific operation.
Deterministic execution
The same contained inputs produce the same reproducible result, every time.
004 / THE EXECUTION LANDSCAPE
Built for a different job.
Browsers optimize pages. Mobile engines optimize apps. Containers and microVMs isolate machines. rayJS is designed around the execution pattern agent platforms actually have.
Landscape descriptions reflect each project's public positioning. rayJS figures are directional internal measurements.
005 / SELECT CUSTOMER ACCESS
Bring us the code you would never let an agent run.
rayJS already supports customer workloads in production. We work directly with agent-platform teams to design safe, high-density execution around their real systems.
Request a rayJS demo →