Get API keyTrusted by 12,000+ developers
One API key, one OpenAI-compatible endpoint, and 250+ models across 40+ providers.
Route, fail over, and see every request without changing your code.
Providers
Forty providers behind one endpoint. Pick model names, not SDKs and secrets.
How it works
Your request arrives once. The gateway picks the route, wears the failures, and returns one response with the full story of how it was served.
Quickstart
Create a key, export it, paste one snippet. The endpoint is OpenAI-compatible, so your existing SDK works unchanged.
curl https://api.airuntime.dev/v1/chat/completions \
-H "Authorization: Bearer $AI_RUNTIME_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "auto",
"messages": [{ "role": "user", "content": "Explain vector search in one paragraph." }]
}'import OpenAI from "openai";
const client = new OpenAI({
apiKey: process.env.AI_RUNTIME_KEY,
baseURL: "https://api.airuntime.dev/v1",
});
const res = await client.chat.completions.create({
model: "auto",
messages: [{ role: "user", content: "Explain vector search in one paragraph." }],
});
console.log(res.choices[0].message.content);import os
from openai import OpenAI
client = OpenAI(
api_key=os.environ["AI_RUNTIME_KEY"],
base_url="https://api.airuntime.dev/v1",
)
res = client.chat.completions.create(
model="auto",
messages=[{"role": "user", "content": "Explain vector search in one paragraph."}],
)
print(res.choices[0].message.content)AI_RUNTIME_KEY.Using your own provider keys instead? Attach them and the router keeps your negotiated rates. Nothing else in the request changes.
Capabilities
Four things that keep working after the demo: routing that adapts, failover that is invisible, caching that pays, and a trace for every call.
Each request is matched to the model that fits its cost, latency, and capability profile. No model list to maintain, no per-provider SDK to wire up.
42msrouting overhead, p50
When a provider degrades, the request is retried against the next best route before your timeout budget is spent. Your users get a response, not a 529.
<80msfailover decision
Identical prompts reuse a stored completion. Cache hits cost nothing and return in single-digit milliseconds, which is where most of the savings actually come from.
38%median cache hit rate
Every call logs the route taken, tokens in and out, cache status, latency, and cost. One trace per request instead of a dashboard you have to reconcile.
1 traceper request
Pricing
Usage is billed per token at the provider rate plus a routing fee. No seat pricing, no minimum, no surprise platform tier.
Enough to ship a side project and find out whether this fits.
$0per month
For production traffic that needs limits it can plan around.
$20per month
For teams with committed spend and a security review to pass.
Customannual
Every plan includes smart routing, automatic failover, and per-request traces. The difference is volume, log retention, and how much of the routing decision you control.
Questions
No. You hold one AI Runtime key and we hold the provider relationships. If you already have your own provider keys, you can attach them and keep your negotiated rates instead.
Yes. Point the OpenAI SDK at our base URL and change nothing else. Model names follow the provider/model convention, so switching a model is a one-line diff and never a rewrite.
The router retries the next best route inside your timeout budget, and the response carries a header telling you which route served it. If every route fails you get one error with the full attempt list, not a silent retry loop.
No. Requests are proxied to the provider you route to and are not used for training. Logs are kept for the window your plan specifies, then deleted, and paid plans can turn logging off entirely.
Per token at the provider rate plus a routing fee, both itemised per request. The free tier is 1M tokens per month with no card. Paid plans bill the included allowance first, then per token.
Yes. Send an explicit model name and the router steps aside. Routing is a default, not a cage: you can pin per request, per endpoint, or per environment.
Start here
Create a key, point your existing OpenAI client at the gateway, and watch the route it takes. A million tokens a month are free, no card.
No card. No sales call. Cancel a plan without emailing anyone.
