Edge AI inference
in under 50 milliseconds
Deploy any model to 300+ edge locations with one command. vr0ng gives you sub-50ms cold starts, autoscaling to zero, and an OpenAI-compatible API, so latency stops being a design constraint.
No credit card · 1M free tokens · deploy in 60 seconds
$ vr0ng deploy ./llama-3-8b
→ snapshotting runtime… done (0.9s)
→ replicating to 300 regions…
✓ live at https://api.vr0ng.online/v1
$ curl .../chat/completions
{ latency_ms: 42, region: "sfo" }
- San Francisco11ms
- Frankfurt9ms
- Singapore14ms
- São Paulo17ms
- Tokyo12ms
Powering inference for fast-moving teams
Everything you need to serve models at the edge
The infrastructure layer for AI products where every millisecond is a conversion. No Kubernetes, no cold-start tax, no ops team.
Sub-50ms cold starts
Snapshot-based runtime restore boots your model in milliseconds, no more waiting on GPU spin-up.
300+ edge regions
Requests hit the location closest to the user. Inference runs where your traffic already is.
Autoscale to zero
Scale from one request to a million and back to nothing. You pay only for the tokens you serve.
Any model, any framework
PyTorch, vLLM, ONNX, GGUF, or your own container. Bring the weights, we handle the rest.
SOC 2 & private by default
Isolated tenancy, encrypted weights, zero-retention logging. Your data never trains anyone else.
OpenAI-compatible API
Point your existing SDK at vr0ng and ship. Drop-in `/v1/chat/completions`, streaming included.
One deploy, everywhere
vr0ng replicates your model to every region automatically and routes each request to the nearest healthy node. No region config, no load balancers to babysit. Traffic just lands where it's fastest.
99.99% uptime
Self-healing nodes and multi-region failover keep you online through any single-region outage.
last 90 days · 0 incidents
Change one line. Keep your code.
vr0ng speaks the OpenAI protocol. Swap the base URL and your existing SDK, prompts, and streaming logic keep working, now served from the edge.
Read the docsimport OpenAI from "openai";
const ai = new OpenAI({
baseURL: "https://api.vr0ng.online/v1",
apiKey: process.env.VR0NG_KEY,
});
const res = await ai.chat.completions.create({
model: "llama-3-8b",
messages: [{ role: "user", content: "hi" }],
stream: true,
});
// first token in 42ms, from the edge“We cut p95 inference latency from 380ms to 44ms the day we moved to vr0ng. Our activation rate jumped 19%. Users can feel the difference.”
Pay for tokens, not idle GPUs
Start free. Scale to zero when idle. No seats, no minimums, no surprise bills.
Hobby
For side projects and prototypes.
- 1M tokens / month free
- 3 edge regions
- OpenAI-compatible API
- Community support
Pro
For teams shipping to production.
- Unlimited tokens, usage-based
- All 300+ edge regions
- Autoscale to zero
- Custom models & containers
- 99.99% uptime SLA
Enterprise
For scale, compliance, and control.
- Dedicated GPU capacity
- Private VPC & peering
- SOC 2 + HIPAA BAA
- Solutions engineer
Ship your model to the edge today
Deploy in 60 seconds with 1M free tokens. Feel the difference latency makes.