A single interface for leading AI models. Every request stays tied to the project or customer behind it.
Same path whether you authenticate with a project API key or a customer token.
Validate the caller, project, customer, and allowed model before provider work begins.
Translate the common request to the model's provider and preserve streaming output.
Record the tokens used and what the provider charged.
Apply usage to the account or customer and enforce the plan that authorized it.
Use the provider-neutral endpoint for new integrations. Existing OpenAI or Anthropic clients can point at the compatible endpoints instead.
curl https://api.modelrelay.ai/api/v1/responses \
-H "Authorization: Bearer $MODELRELAY_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model":"claude-sonnet-5","input":"Hello"}'Change the model. Keep everything else.
Control which models each customer tier can use and designate the default explicitly.
Give your backend a customer-scoped credential so requests resolve to the correct end user.
Enforce subscription limits or PAYGO balances before spending reaches the provider.
ModelRelay sits in the request path, so model choice, access policy, and billing stay one system instead of three.
| Direct provider integration | ModelRelay | |
|---|---|---|
| Model access | Provider-specific clients | Every model through one API |
| Customer identity | Build the mapping | Customer-scoped tokens |
| Usage controls | Build per provider | Tiers, limits, and balances |
| Customer billing | Reconcile separately | Usage ledger plus Stripe-backed plans |
Open to everyone. Every new account starts with $1 in credit.