Your users pay for inference. You should see some of it.
If your app asks people to bring their own API key, you carry the support burden, write the integration docs and drive the spend — and earn nothing on any of it. We pay 10% of everything your users spend with us, credited daily.
No minimum volume. No exclusivity. No contract to sign before you know it works.
Three steps, one of which you have probably already done
Attribution rides on X-Title, the same header apps already send to identify themselves to OpenRouter. If yours sends it, the integration work is finished.
Send your app name
One header on the requests your users already make. Nothing about your auth, billing or user data changes — they keep their own key and their own account.
We link it to your account
Done by hand, deliberately. Attribution headers are caller-controlled, so a human confirming “this app belongs to this account” is the only join we will pay against.
Credits land daily
Each completed day is credited once, and shows in your transactions as a line you can reconcile — the app, the percentage, the spend it came from, and the date.
POST https://api.infersia.com/v1/chat/completions
Authorization: Bearer isk-v1-...
X-Title: YourApp
{ "model": "deepseek/deepseek-v4-flash-0731", "messages": [...] }A rebate is only worth having if you can recommend us honestly
Pointing your users at a provider you would not use yourself costs more than it pays. So here is what they actually get, and why we think it stands up.
The full context window, not a slice
We serve DeepSeek V4 Flash at its complete 1,048,576 tokens. Most pools cap the same model at 262,144 — a difference your users hit as a hard error, not a slowdown. If your product involves whole codebases, long documents or agent sessions that accumulate, that ceiling is the thing they complain about.
We publish the quantisation, so your benchmarks mean something
Every model states the precision it is served at, the hardware behind it, and measured latency and uptime. Quantisation is the biggest hidden variable in commercial inference and almost nobody discloses it — which is why the same model behaves differently across providers and your users blame your app. How we measure it.
A free tier, so trying costs them nothing
qwen/qwen3-8b:free runs at no cost with no card, which removes the payment step from your onboarding as much as ours. A new user can finish your tutorial before deciding to pay anyone.
It already works in your app
OpenAI-compatible, verified against the official SDKs — streaming, tool calls, structured output and prefix caching. If your app accepts a custom base URL, your users can point at us today without you shipping anything.
The parts people email to ask about
- What exactly is 10% of?
- Of what your users are actually billed — the same figure on their invoice, after any discount or promotional rate. Not list price, and not a margin we compute privately.
- Credits, or cash?
- Credits on your Infersia account, which is the honest framing: it is worth most to a partner who also builds on us. If your volume makes that the wrong shape, say so in your first email — the percentage and the mechanics are both negotiable.
- Does my own usage count?
- Yes. Every request attributed to your app counts, including your own testing. We would rather say that plainly than have you discover it and wonder what else went unsaid.
- What if we stop?
- Credits already granted stay granted. The ledger is append-only and ending a partnership is not a clawback. There is no lock-in, no exclusivity, and nothing to unwind.
- Is my users’ data involved?
- No. We hold no prompts or completions, so there is nothing to share even if someone asked. All you send is the name of your app; all we send back is a number.
Tell us what you’ve built
A link and a sentence is enough to start. We will tell you within a day whether it fits, what percentage we can do, and what — if anything — you would need to change.
Early stage and small is fine. We would rather back something growing than wait for it to arrive already large.