Question 1

What is an AI API gateway?

Accepted Answer

An AI API gateway is a single endpoint that routes your requests to many upstream model providers. Instead of integrating OpenAI, Claude, Gemini, and others separately, you call one OpenAI-compatible API and the gateway handles authentication, protocol translation, load balancing, and billing.

Question 2

Which AI models and providers are supported?

Accepted Answer

It aggregates 40+ upstream providers, including OpenAI, Anthropic Claude, Google Gemini, Azure OpenAI, AWS Bedrock, DeepSeek, and Qwen, behind one unified protocol. New channels can be added as providers ship.

Question 3

Is it OpenAI API compatible?

Accepted Answer

Yes. Requests and responses follow the OpenAI-compatible format, so existing SDKs and tools work by changing only the base URL and API key. Native Claude and Gemini endpoints are also exposed for clients that need them.

Question 4

How is usage billed?

Accepted Answer

Billing is pay-as-you-go and transparent: every request records prompt and completion tokens, latency, and cost. You can review usage in real time, set quotas and rate limits per user or token, and see the cost of each request before you commit.

Question 5

Can I self-host the gateway?

Accepted Answer

Yes. The gateway is self-hostable and open source, with SQLite, MySQL, or PostgreSQL for the primary database and optional Redis and ClickHouse for caching and logs. You keep control of your keys, data, and deployment.

Question 6

How does it handle reliability and failover?

Accepted Answer

Requests are routed to the healthiest upstream based on live channel status. Unhealthy channels are taken out of rotation automatically, and load balancing spreads traffic across the providers you configure so a single outage does not take you down.

听风API: one API endpoint for many AI models

Limitations