wide area network → wide area intelligence

Your hardware is
the datacenter.
The cloud is the failover.

Wide Area Intelligence is an OpenAI-compatible AI gateway that routes every request to your own hardware first — office workstations, on-prem servers, lab machines — and silently fails over to the cloud when it has to. It's an LLM gateway for the models you self-host: your apps never know the difference, and requests served by your own hardware carry no per-token fees.

[ 2 nodes free ][ openai-compatible ][ no port forwarding ]
live route topology routing
YOURAPPWIDE AREAGATEWAYcache · route · failoverYOUR GPUno token feesCLOUDfailover onlyopenai sdkPRIORITY 1 · tunnelPRIORITY 2 · openrouter
dstviametricstate
inference/*node-01.tunnel0UP
inference/*openrouter.ai100STANDBY
cache/hitedge.kv38.2%
$0.00
per-token fees on your own hardware
<50ms
edge routing overhead
3 hops
cache → local → cloud
100%
openai sdk compatible

/// the explainer

The idea, in a couple of minutes.

What Wide Area Intelligence is, and why running inference on your own hardware first — cache, then your GPUs, then the cloud as plan B — costs less than sending every token to someone else's API.

explainer.mp4
A quick explainer — what it is and how hybrid routing works.

/// powering inference across the portfolio

/// what brings you here

One gateway. Three reasons to run it.

/// how it works

The AI gateway, in three tiers. Zero code changes.

01metric 0 · ~10ms

EDGE CACHE

Identical request seen before? Served instantly from Cloudflare's edge — the prompt never touches a GPU. Toggle it per-account, set your own TTL.

02metric 1 · no token fees

YOUR HARDWARE

A one-line Docker command turns any machine with a GPU into a node. It opens a secure Cloudflare Tunnel — no port forwarding, no static IP — and serves llama.cpp behind your gateway.

03metric 2 · always up

CLOUD FAILOVER

Node offline? Timed out? The same request silently re-routes to the cloud — GPT, Gemini, Claude — through our managed gateway, billed only against your prepaid credits. If one provider has an outage, the gateway fails over to the next. Your app sees one response, every time.

/// node setup

One command.
Any machine. Any OS.

The installer finds Docker, Podman, or Colima — or sets one up for you — then starts an agent bundling llama.cpp and a Cloudflare Tunnel. Run it on a workstation, a Mac Studio, or a rack server in the office: it registers itself and starts serving traffic in about a minute.

node-01 · office
$ curl -fsSL https://wideareaai.com/install.sh \
    | sh -s -- --key wai_node_…

[wai] Platform: linux
[wai] Using container runtime: docker
[wai] NVIDIA GPU passthrough enabled
✓ tunnel established
✓ node registered · node-01 ONLINE

/// app integration

Keep your SDK.
Change one URL.

Anything that speaks the OpenAI API speaks Wide Area Intelligence — Python, TypeScript, LangChain, agents, curl. Point it at the gateway and routing, caching, and failover happen automatically.

app.py
from openai import OpenAI

client = OpenAI(
    base_url="https://wideareaai.com/api/v1",
    api_key="wai_sk_…",
)

resp = client.chat.completions.create(
    model="llama-3.1-8b-instruct",
    messages=[…],
)

/// pricing

Own your baseline. Burst to the cloud.

Your GPUs serve the baseline with no per-token fees — your costs are the nodes and the power bill. The first two nodes are free; each node after that is $100/mo ($1000/yr). Team seats are unlimited. Cloud failover is prepaid credits, billed per request only on the overflow.

STARTER

$0forever
  • 2 compute nodes — free
  • Unlimited team seats
  • 2M requests / mo per node
  • Exact-match edge cache
  • Cloud failover (prepaid credits)
Start free
most popular

PER NODE

$100/ mo per node
  • Everything in Starter
  • Each node beyond the first 2
  • Or $1000 / yr per node (2 months free)
  • Auto-billed as you add & remove nodes
  • Cloud usage still metered on credits
Add a node

/// initialize

Your hardware is already paid for.
Put it on the network.

Create your gateway →