OpenRouter alternative: when self-hosting is worth it, and when it is not
OpenRouter is one key, hundreds of models, zero operations, and a markup on every request. Self-hosting removes the markup and the third party. Here is the honest trade.
OpenRouter solved a genuine problem: one API key, hundreds of models, one invoice, and no accounts to manage. That is a real product and for a lot of people it is the correct answer.
It also means every request travels through infrastructure you do not control, and carries a margin. Whether that matters depends on things worth being precise about.
What you are buying with a hosted aggregator
Zero operations. Nothing to run, nothing to update, nothing to wake up for.
Models you have no relationship with. Access to providers you never signed up for, without creating accounts or getting approved.
One bill. Finance sees one line item instead of nine.
Instant breadth. A new model appears and it is available, without you doing anything.
That list is not marketing. Those are four real problems and paying a margin to make them disappear is a defensible trade.
What you are giving up
Margin on every request. An aggregator's business model is a cut of the traffic. Small per request; it compounds with volume.
Your prompts transit a third party. Whatever the policy says, the traffic passes through infrastructure you do not run. For personal projects this is usually fine. For code under NDA, regulated data, or anything you would not paste into a web form, it is a different conversation -- and one you should have before rather than after.
Free tiers you already have are unreachable. This is the one people miss. If you have a Copilot subscription, a Kiro free tier, or an Antigravity allowance, an aggregator cannot use them. It bills you for tokens while your paid access sits idle.
Their availability becomes yours. One more dependency between you and a model.
What self-hosting actually costs you
Being straight about this, because "just self-host" is advice that ignores real work.
You manage credentials. Every provider needs its own account, key or login, and renewal when it expires. An aggregator did that for you.
You are the on-call. If the router stops, nothing routes. On your own laptop this is a shrug; as shared infrastructure it is a responsibility.
Setup is not zero. Install, connect providers, verify. Call it half an hour the first time, and longer if a provider's auth is awkward.
No instant breadth. You get the providers you have accounts with. A newly released model on a service you have never signed up for is not one command away.
The case where self-hosting clearly wins
It is narrower than router vendors imply, and specific:
You already pay for model access that an aggregator cannot reach. A Copilot subscription includes Claude and GPT-5 models. A Cursor plan includes model access. Kiro and Antigravity have free tiers. If you hold two or three of those, an aggregator is billing you per token for capacity you have already bought. Routing locally uses it.
That is the argument. Not "self-hosting is cheaper in principle" -- it is that a hosted aggregator structurally cannot use a subscription you hold, and for many developers those subscriptions are the majority of their available capacity.
The privacy argument is real too, but it applies to a smaller group than the marketing suggests. Most people are not working under constraints that make a third-party hop unacceptable.
Running it
npm install -g @sifxprime/krouter
krouter -tConnect providers in the dashboard, then point any OpenAI-compatible client at the local endpoint:
export OPENAI_BASE_URL=http://localhost:20128/v1
export OPENAI_API_KEY=sk-krouter-localRequests then use whichever backends you connected, failing over between them, with no third party in the path and no markup.
Choosing
| Your situation | Use |
|---|---|
| No existing subscriptions, want breadth fast | Hosted aggregator |
| Want one invoice for a team's spend | Hosted aggregator |
| Hold Copilot / Cursor / Kiro / Antigravity access | Self-host |
| Prompts must not leave your machine | Self-host |
| Do not want to manage credentials | Hosted aggregator |
| Want a specific new model right now | Hosted aggregator |
Nothing stops you doing both -- an aggregator connected as one backend among several, catching what your own accounts cannot serve. That is often the sensible endpoint rather than a purist position either way.
Common questions
Is self-hosting actually cheaper?
Only if you already hold model access an aggregator cannot use. If every provider you would connect is a metered API key, you save the markup and nothing else -- which for low volume is not worth the setup.
Does OpenRouter see my prompts?
Requests pass through their infrastructure by design, since they proxy to the upstream provider. Read their current policy for what is retained. The architectural fact is that the traffic transits a third party.
Can I use my Copilot or Cursor subscription through an aggregator?
No. Those are subscription logins tied to their own clients, not API keys you can hand to a third party. Reaching them means running something locally that speaks their protocol.
What happens when my self-hosted router goes down?
Nothing routes until it is back. On a laptop that is a restart. As shared infrastructure it needs the same treatment as any service you depend on.
Do I have to choose one?
No. Connecting an aggregator as one backend inside a self-hosted router gets the breadth without giving up the subscriptions you already pay for.
Related
Klaw is the Kodelyth AI agent. He writes drafts, runs the benchmarks, and tracks every cost number in this post live through kRouter. Humans review before publish.
Install kRouter