Sell your inference through Hopscotch
Hopscotch gives developers one API for models from many providers. List your models once, and customers can choose you by name or add you as a fallback.
What you get on the network
One integration puts your models in front of developers who already call Hopscotch.
One listing, many customers
Your models sit alongside the rest of the catalog. Developers call you by name, with nothing new to integrate.
Chosen directly or as a fallback
Beyond customers who choose you directly, you can be set as someone's fallback provider.
Every API type you serve
Chat Completions and Responses today. Tell us what else you run when you onboard.
Latency tracked on real requests
We measure latency from live traffic, so customers compare providers on what actually happens, not on claims.
Region and data rules enforced
Declare where you serve from and whether you offer zero data retention. Requests that require either only reach providers that meet it.
Capacity limits respected
Return a 429 when you're full and Hopscotch moves the request on, instead of piling more onto a busy endpoint.
What we look for
You don't need all of these to apply. Switch on what you have to see how close you are to going live.
Your inference API streams responses
Any format we support works.
Usage comes back on every response
Token counts on streamed and non-streamed responses alike. They're how requests are metered.
You publish a models endpoint
Each model with its pricing, context length, max output, features and regions. If you already sell through another router, yours will likely work as is.
You publish a privacy policy and terms
Saying whether prompts are logged, how long data is kept, and whether it's used for training.
From application to live traffic
Most of the work happens on our side once you've applied.
Apply
Your company, your models, and how you sell today. About two minutes.
Review
We look at your models, your pricing and where you'd fill a gap in the catalog, then reply to your contact email.
Technical onboarding
Your answers carry over. An AI assistant can draft the API details, and we test your endpoints with real traffic.
Go live
They choose you by name, or add you as a fallback.
Questions providers ask
What API format do I need?
Any API that streams responses and reports usage tokens. OpenAI-compatible /chat/completions is the quickest to connect.
How do customers end up using my models?
Two ways: they choose you directly, or they set you as a fallback. Making your models fast, well priced, and clearly described helps with both.
How is pricing set?
You set your own prices, per model, in your models endpoint. Keep it current, since that's where we read your prices. Your base price should be your lowest rate.
Can I charge different prices for different models or tiers?
Yes. Pricing is per model, and if you run a standard and a fast tier of the same model, each can have its own price, quantization, and hardware.
How are region and data requirements handled?
You declare your regions and whether you offer zero data retention. Requests that require a specific region or ZDR are only sent to providers that meet the requirement.
Do I have to support every feature?
No. List what each model supports in your models endpoint so customers can see it.
How do I get paid?
Usage is metered from the token counts your API reports. We'll agree payment terms with you during onboarding.
Can I apply if I don't meet every requirement yet?
Yes. Apply now and tell us where you are. We'll work through the gaps with you during technical onboarding.
Ready to apply?
We read every application. Providers with models or strengths we don't have yet, like proprietary models, unusual hardware, or workload specialties, are reviewed first.
Something else? Tell us in the last box of the application, or write to partnerships@hopscotchlabs.ai.