Start with the model gateway you already run
ModelVendor is built for an operator who already runs a model server. You do not move that model into ModelVendor or rebuild its serving stack. Instead, you keep the model gateway in the environment where you operate it and run the ModelVendor connector alongside it. The connector is the bridge between the marketplace and your existing OpenAI-compatible gateway.
Before you connect, make sure your gateway can answer GET /v1/models. ModelVendor uses that response to discover the models your server currently advertises. This matters because a seller hostname is not a claim that every model is available forever. The connector reports what the gateway exposes, and the seller flow uses that information when you choose what to list.
Create a connection token and run the connector
Open the seller flow at /sell and create a private connection token. Treat that token like a credential for this connector: it attaches the gateway to your account, and issuing a replacement token invalidates the earlier one. Install the connector on the machine that can reach your model gateway, then run it with the token and your gateway URL as the upstream.
The connector stays running with your model server. When it authenticates, ModelVendor provisions the seller hostname for the account. The seller does not need to type a temporary tunnel URL or manually manage certificates in this flow. Your model server continues to run where you operate it; the connector supplies the marketplace connection around it.
Use one connector for more than one model
A single gateway can advertise several models, and one connector can expose those models. That does not force every model into one listing. Each model can have its own marketplace listing, capacity settings, and price. This is useful when one machine serves several models or when a gateway aggregates capacity behind one OpenAI-compatible endpoint.
Choose the models you want to list after the connector has discovered them. If a model is not shown, first check the exact name returned by your gateway's GET /v1/models response. The seller flow also provides a place to enter an exact API model name when discovery does not surface the name you need.
Set prices on a consistent token unit
For each listing, set separate prices per million input tokens and output tokens. ModelVendor does not infer those prices for you. Buyers are billed from the request and response that ModelVendor relays, using input and output token counts computed from the response actually forwarded. The listing price therefore has two parts: input usage and output usage.
ModelVendor uses its versioned billing tokenizer so listed models are counted on a consistent unit. This is designed to make pricing comparable across listings without relying on a seller-supplied usage field. It does not remove the operator's responsibility to choose a price that fits the model and capacity they are offering.
Your next step
Go to /sell, create a connection token, point the connector at the OpenAI-compatible gateway you already run, then select a discovered model and set its input and output token prices. If you want help verifying the first connection, the seller flow includes a direct way to schedule an onboarding call with the founder.
