Build with Inferbase
Bring a production AI workload and shape what we build next. Design partners get Pro at no charge, inference credits and direct support from the people who build it.
What you get, and what we ask
We work closely with a small number of teams at a time. You get the platform and our time; we learn from a real workload what to build next.
What you get
- Pro at no charge for the length of the program
- Inference credits for the managed models
- Direct support from the engineers who build Inferbase
- Onboarding done with you: keys, cloud accounts, routing policy, guardrails and budgets
- A say in what we build next, with your requests first in the queue
What we ask
- A real workload: one feature in production, or close to it, running through Inferbase
- A 30-minute call every two weeks
- Candid feedback, most of all on what does not work
- Permission to name you as a design partner, once you are happy with it
Who it suits
Teams that ship LLM features to their own customers. It fits best when one or more of these is true today.
Several models
You call more than one model or provider, and choosing between them is a decision someone makes by hand.
Failures cost you
When a provider times out or rate-limits, your users notice. You want the request answered by the next model, not retried by your code.
Your cloud, your keys
Data has to stay in your own Azure, Bedrock or Vertex account, or on your own provider keys, and you still want one API in front of it.
Shared budgets
Several teams, services or customers draw on one AI budget, and you need limits per key and per project rather than one bill at month end.
How it works
Four steps from application to real traffic on one feature.
- 01ApplyTwo minutes. Tell us what you are building and how you run models today.
- 02Intro callThirty minutes on your current setup. You decide whether it fits.
- 03OnboardWe set up the project, keys, policy, budgets and any cloud accounts with you.
- 04Run itReal traffic on one feature first, and a call every two weeks on how it goes.
Apply to the program
A few questions so the first call starts from your setup rather than from ours.
Frequently asked questions
What the program costs, who it suits, what happens to your data, and what follows an application.
A small group of teams that bring a production AI workload to Inferbase and work with us closely while they run it. You get the platform, inference credits and direct access to the people who build it. We get to watch it carry real traffic and hear what is missing, which decides what we build next.
No. Design partners are on Pro at no charge for the length of the program, and receive inference credits for managed models. Requests on your own provider keys or cloud accounts are billed by your provider, as they would be without Inferbase.
Teams that ship LLM features to their own customers, or are about to. It suits you best when you call more than one model or provider, need a fallback when one fails, need data to stay in your own cloud account, or share one AI budget across several teams or customers.
No. Most teams start with one feature or one service. Inferbase is an OpenAI-compatible API, so moving a service is usually a change of base URL and key, and moving it back is the same change.
Yes. Connect your OpenAI, Anthropic, Gemini and other provider keys, or your Azure, AWS Bedrock or Google Vertex account, and requests run on them. Your provider bills the tokens and Inferbase adds nothing to them. You can also mix your own keys with the managed catalog in one project.
Each API key has a setting for whether its request content is kept in the 30-day request log. Turn it off and nothing from that key is stored. A workspace or user zero-retention setting applies on top of it. Metadata such as the model, tokens, latency and cost is always recorded, because that is what the logs and budgets are built from.
With your permission, and only once you are happy with how it works, we list your company as a design partner, for example on this site or in a conversation with an investor. Nothing is published without your approval of the exact wording, and you can say no to this and still join.
We read every application and reply within two business days. If it looks like a fit, we set up a 30-minute call to look at your current setup. If you decide to go ahead, we onboard you together: the project, keys, routing policy, guardrails, budgets and any cloud accounts.