AI API gateway
Use InferGate as the API entry point for model requests, key management, prepaid balance and usage records in server-side applications.
An API gateway gives your application one entry point while keeping account access and request usage visible. InferGate exposes a compatible model API on the API host and account controls on the application host.
Three hosts, clear responsibilities
- useinfergate.com: public documentation, models and integration guides.
- app.useinfergate.com: account, keys, wallet and usage screens.
- api.useinfergate.com/v1: authenticated model requests.
Application responsibilities
Keep secrets server-side, define timeouts, handle stream completion, and review request outcomes. A gateway does not remove the need for your own workload testing or guarantee that an upstream provider will never fail.
From account to first response
- Create an account or sign in to your existing account.
- Check wallet balance and create a server-side API key.
- List the models your key can access, then send one small request.
- Inspect the returned model, completion status and usage before increasing traffic.
Start with an integration
Gateway and application responsibilities
| Concern | InferGate surface | Your application responsibility |
|---|---|---|
| Access | API keys and account permissions | Store and rotate secrets securely |
| Requests | One API base URL and published models | Choose explicit IDs and validate results |
| Usage | Account balance and usage records | Estimate workload costs and reconcile usage |
| Reliability | Request status and returned errors | Set timeouts and handle interruptions |
Runnable integration example
curl --fail-with-body https://api.useinfergate.com/v1/responses \
-H "Authorization: Bearer $INFERGATE_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model":"gpt-5.5","input":"Return a short release note for an API client update.","max_output_tokens":256}'Frequently asked questions
Do I need to change my application architecture?
A server-side SDK client can usually be configured with a separate base URL and key. Keep that change behind your existing client boundary and verify actual request behavior before migrating traffic.
Does a gateway eliminate upstream failures?
No. Clients still need timeout, cancellation and error handling. Record failures separately from completed responses and do not blindly retry partially consumed streams.
How do I choose between the model variants?
Use a representative evaluation set with expected answers, completion checks, observed latency and actual usage cost. There is no universal best variant asserted here.
Continue your integration
Review prepaid pricing · Read API documentation · Get Started