the ai gateway for hardware you own
An AI gateway that routes to GPUs you own first.
Wide Area Intelligence is an OpenAI-compatible AI gateway — one endpoint in front of every model and provider. It serves repeated requests from an edge cache, runs the rest on hardware you already own with no per-token fees, and bursts to the cloud only when a node can't serve. Change one line of code and your apps route themselves.
/// definition
What is an AI gateway?
An AI gateway is a single API endpoint that sits in front of every large language model and provider an application uses. Instead of each app talking directly to OpenAI, Anthropic, or a local inference server, every call goes through the gateway — and the gateway owns the cross-cutting concerns: routing requests to the best available model, caching repeated responses, failing over when a backend is down, and enforcing keys, budgets, and rate limits in one place.
Wide Area Intelligence is an own-hardware-first AI gateway. The routing order is the whole product: an edge cache answers repeats in milliseconds, GPUs you own answer the rest with no per-token fees, and the cloud is burst-only failover. Your apps see one stable, OpenAI-compatible endpoint; the gateway decides where each request actually runs.
/// the trade-offs
AI gateway vs. the alternatives
A raw provider API locks you to one vendor. A cloud aggregator like OpenRouter still bills every token and ships your data off-network. Running models directly is free but reachable only on your LAN. An own-hardware-first AI gateway is the combination that doesn't force the trade.
| Approach | Routing | Baseline cost | Failover | Your data |
|---|---|---|---|---|
| Raw provider API | One vendor, one key | Every token billed | None — vendor outage = your outage | Leaves your network |
| OpenRouter / cloud aggregator | Many cloud models, one key | Every token billed (+ markup) | Across cloud providers only | Leaves your network |
| Running models directly | Your box only, on your LAN | No token fees | None — box down = no answer | Stays local (LAN-only reach) |
| Wide Area Intelligence | Cache → your GPUs → cloud | No per-token fees on your hardware | Cloud burst on overflow, auto | Yours — leaves only on burst |
/// how the routing works
One AI gateway, three tiers. Cache the repeats, own the baseline, burst the rest.
EDGE CACHE
An identical request you've served before is returned straight from Cloudflare's edge — the prompt never reaches a GPU. Per-account toggle, your own TTL.
YOUR HARDWARE
A one-line install 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 with no per-token fees.
CLOUD FAILOVER
Node busy, offline, or timed out? The same request silently re-routes to the cloud through our managed gateway, billed only against prepaid credits — burst only, never the baseline.
/// drop-in integration
Point your OpenAI SDK
at the gateway.
Anything that already speaks the OpenAI API speaks Wide Area Intelligence — Python, TypeScript, LangChain, agents, curl. Swap the base URL for the gateway and routing, caching, and cloud failover happen automatically. No SDK swap, no new client.
- Change one line: the base URL.
- Works with the OpenAI SDK, LangChain, agents, curl — anything OpenAI-compatible.
- Bring your own gateway key; routing, caching, and failover are automatic.
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=[…], )
/// why this gateway and not another
Own your baseline
Steady-state traffic runs on GPUs you already paid for, with no per-token fees. Your marginal cost is the power bill, not a metered API — and the cloud only ever catches the overflow.
OpenAI-compatible, drop-in
One endpoint, the API your stack already speaks. Change the base URL and existing code, agents, and SDKs route through the gateway unchanged.
Yours, not theirs
Requests run on your hardware and your data leaves only when you choose to burst. The gateway routes on Cloudflare's edge; compute stays on machines you control.
/// go deeper
Plan your gateway
New to the concept? Start with what is an LLM gateway, compare options in the best OpenRouter alternatives, or browse the model directory.
/// faq
AI gateway questions
- What is an AI gateway?
- An AI gateway is a single API endpoint that sits in front of every large language model and provider you use. Instead of wiring each app directly to OpenAI, Anthropic, or a local server, your code calls the gateway and the gateway handles routing, caching, failover, rate limiting, key management, and cost control. Wide Area Intelligence is an OpenAI-compatible AI gateway that routes each request to GPUs you own first and bursts to the cloud only when it has to.
- How is an AI gateway different from calling the provider API directly?
- Calling a provider directly couples every app to one vendor, one key, and one set of rate limits — and bills every token. An AI gateway decouples your apps from providers: you change one base URL, and routing, model fallback, edge caching of repeated requests, and budget enforcement happen in one place. With Wide Area Intelligence the baseline runs on your own hardware with no per-token fees, and the cloud is reserved for overflow.
- Is an AI gateway the same as OpenRouter?
- OpenRouter is a cloud aggregator: every request still leaves your network and bills per token across third-party providers. Wide Area Intelligence is an own-hardware-first AI gateway — it routes to GPUs you control first, caches repeats at the edge, and only fails over to the cloud, so your steady-state requests carry no per-token markup and your data leaves only when you send it.
- Do I have to rewrite my code to use an AI gateway?
- No. Wide Area Intelligence speaks the OpenAI API, so anything that already uses the OpenAI SDK, LangChain, an agent framework, or curl works by changing a single line — the base URL. No SDK swap, no new client library.
- What does an AI gateway cost?
- Wide Area Intelligence starts free with 2 nodes and 2M requests/month per node. Requests served from your own GPUs carry no per-token fees — your only marginal cost is electricity. Cloud failover is prepaid credits, billed per request only on the overflow, never on the baseline.
/// initialize
Put every model behind one endpoint. Route to your own GPUs first.
no credit card · 2 nodes free, forever · openai-compatible