Requirements
Your server must:- Speak the OpenAI API. TokenChange sends streaming requests to
/v1/chat/completionsand/v1/completionsand checks/v1/models. - List the model.
/v1/modelsmust include the model name your server expects, which defaults to the catalog slug. - Stream with usage. Responses must be server-sent events. The final usage chunk must report
prompt_tokens,completion_tokensandprompt_tokens_details.cached_tokens; billing depends on these numbers. - Be reachable on the public internet over HTTPS at a public DNS name. Private, internal and single-label hostnames are refused.
Authorization: Bearer ....
The model must already be in the catalog. Only open-weights models are listed.
How it works
1
Add an endpoint
Open Provider > Endpoints and select Add endpoint. Your first endpoint also creates your provider account. See Add an endpoint.
2
Operator review
A TokenChange operator reviews every endpoint you add. Your provider account is approved together with your first endpoint.
3
Validation
Once approved, TokenChange tests your endpoint automatically. It goes into rotation when every required check passes. See Review and validation.
4
Serve and earn
Customer requests reach your endpoint up to its slot limit. Each request earns you credits at your prices, minus the platform fee. See Pricing and earnings.
Limits
- Up to 20 endpoints per provider account. Delete one to add another.
- TokenChange never sends an endpoint more concurrent requests than its slots.
Suspended provider accounts
If TokenChange suspends your provider account, your endpoints leave rotation and become read-only. You can still view and delete them. The banner on the Endpoints page shows the reason; contact support@tokenchange.io to resolve it. A provider suspension doesn’t affect your access as a customer.Next steps
Add an endpoint
Every field in the endpoint form.
Pricing and earnings
How you set prices and get paid.