What a relay does, how to judge one, and how ours works.
An AI API relay forwards your requests to upstream model providers and returns the response, so you get one endpoint, one key and one bill across several providers. This page covers what that means in practice and what separates a reliable relay from a fragile one.
A relay sits between your client and upstream model providers. You send an OpenAI- or Anthropic-style request to the relay, it picks a healthy upstream channel, forwards the request, and streams the response back. Your code keeps its existing SDK; only the base URL and key change.
Four things matter more than price. Does it publish live channel status instead of asking you to trust it? Does it fail over automatically when an upstream degrades? Does it log tokens and cost per request so you can audit the bill? And does it support the native request shapes — /v1/responses and /v1/messages — rather than forcing everything through chat completions?
Claude Code, Codex, GPT, Gemini and Grok behind one endpoint, with a public group status board, automatic failover across channels in a group, per-request usage logs showing model, tokens and cost, and plans billed by day, week or month.
In practice they describe the same thing: a service that forwards model API requests upstream and returns the response. A relay usually adds key management, quotas, billing and routing on top of plain forwarding.
Usually only the base URL and API key. Most OpenAI and Anthropic SDKs, plus tools like Claude Code and the Codex CLI, work unchanged after that config change.
Check whether it publishes a live status board for each channel group, whether it documents failover behavior, and whether usage logs let you reconcile every request against the amount billed.
Claude and Claude Code, OpenAI Codex and GPT, Google Gemini, and xAI Grok — all through a single key and endpoint, switching by model name per request.
Keep Codex running on a stable endpoint you control.
Claude Code API & Claude API gatewayStable Anthropic Messages access for Claude Code and agents.
GPT API & OpenAI Responses gatewayLatest GPT, Responses and Chat Completions through one endpoint.
OpenAI-compatible API gatewayOne endpoint, many models, automatic failover.