Every frontier model. One key.
Cost + 5% — or half price when you can wait.
Point your existing OpenAI SDK at one endpoint and route GPT, Claude, image and video models through a single prepaid wallet. Urgent work runs now at the raw provider price plus a flat 5%. Anything that can wait an hour runs on the provider's own 50%-off Batch API — same models, half the list price.
Change one line, keep your stack
The gateway speaks the OpenAI API. Swap the base URL and every request comes back with a receipt: the provider's raw price, your +5%, and the exact routing trail — down to the nano-dollar.
from openai import OpenAI
client = OpenAI(
base_url="https://gateway.interwebs.ai/api/v1",
api_key="iwk_live_…", # minted in your dashboard
)
r = client.chat.completions.create(
model="auto", # cheapest live model that clears standard quality
messages=[{"role": "user", "content": "Summarise this contract…"}],
)
print(r.model_extra["interwebs"])
# → { "raw_cost_nano": …, # the provider's real price
# "charged_usd": …, # you: raw + exactly 5%
# "skim_bps": 500, # our whole fee
# "attempts": […] } # the routing trailWhat's happening under the hood
Every request is priced off a live board of real provider rates. The cheapest live model that clears your quality bar wins the request; if it fails, routing escalates to the next one. This is that board — real prices, real fills, nothing simulated.
From zero to your first request in three steps
Load your wallet
Create an account and top up by card via Stripe Checkout. No subscription, no minimum — unused balance is refundable at any time, and it only meters down as requests bill.
Mint scoped keys
Create keys in one click from the dashboard. Name them, cap them with per-job and per-period spend mandates, and revoke them instantly — safe defaults for autonomous agents.
Swap the base URL
Keep your existing OpenAI SDK and change one line. Ask for a model by name, a quality tier like auto:premium, or one of your own custom routers.
Five modalities, one balance
Text, image, video, PDF and audio behind a single gateway and a single wallet. Swap between GPT, Claude, DeepSeek, FLUX, Veo and more by changing one string.
Metered to the nano-dollar
An append-only ledger records every request, so your spend is exact, idempotent and auditable.
Price-aware routing
Ask for auto:economy, auto:standard or auto:premium and each request routes to the cheapest live model that clears the bar — re-priced from the live board on every call.
Automatic failover
If a provider fails mid-request, routing escalates to the next-cheapest model that clears your quality bar, and the attempt trail lands on your receipt. No output, no charge.
Built for agents
Machine-to-machine keys with spend mandates and off-session auto-reload — an autonomous agent can spend unattended without ever exceeding the budget you set.
Receipts, not promises
Every response shows the raw provider price next to what you paid. Batch fills settle at the provider’s real discounted rate — never a made-up number.
Refer & earn — 2% + 1%, for life
Every account has a permanent referral link. Put it in your README, your app's onboarding, your agent's docs — anyone who signs up through it earns you 2% of everything they ever spend, plus 1% from the people they refer.
Out of our fee, not their pocket
Commission is carved out of our 5%. Your users still pay raw cost + 5% — never a penny more — so recommending the cheapest gateway costs them nothing.
Settles as it bills
No net-30, no payout thresholds. Commission lands in your wallet as spendable balance the moment a referred request bills — and runs your own workloads on any model.
Made for tools that spread
Open-source packages, wrappers, agent templates — whatever you ship, the accounts it brings in earn for you permanently. Your tool can even onboard its users machine-to-machine with one API call. Attribution is set once at signup and can never be hijacked downstream.
Built on the gateway
Two reference apps run on the same public API you get — proof, not products.
Ask a question, get a cited PDF report
A planner breaks your question into steps, routes each to the best-priced model that clears the bar, runs them in parallel and assembles a referenced PDF — with a per-step receipt showing what ran, what it cost, and what the routing saved.
Run a research job →A no-code playground for the visual models
Generate with Veo, Sora, Kling, FLUX, GPT Image and more from the dashboard — no API key needed. The same real provider rate plus 5%, with the price shown before you hit Generate. Try outputs before you write a line of code.
Open the Studio →Ship yours — and let it pay you
Build on the gateway, put your referral link in your README or onboarding, and earn 2% of everything your users ever spend — carved out of our fee, not their pocket.
See refer & earn →Two lanes, one honest rule
Every request is billed at the raw provider price plus a flat 5% — no seats, no tiers, no minimums, no monthly bill. The spot lane runs now at cost + 5%. The batch lane runs async on the provider's own Batch API at 50% off list, plus the same 5%. Your receipt shows raw vs charged on every request.
Mint a key and make your first call in minutes
Mint a free API keyAlready have an account? Log in