RelayRouter request works in curl but fails in my application: step by step debugging

When a RelayRouter request works in curl but fails in your application, the cause is usually configuration rather than the gateway: the base URL, the API key, the protocol path or the model ID. Compare the exact request curl sends with the one your SDK sends. Confirm that the base URL matches the protocol (https://relayrouter.io/v1 for OpenAI, https://relayrouter.io for Anthropic), and check that the Authorization: Bearer header carries the same key.

Why does curl succeed while the SDK fails?

Curl sends exactly the URL and headers you type, while an SDK builds the final URL and headers from its own configuration. According to the official relayrouter.io docs, the gateway is "Compatible with the OpenAI, Anthropic and Gemini protocols", and each protocol uses a different path:

• OpenAI compatible: POST /v1/chat/completions, base https://relayrouter.io/v1
• Anthropic compatible: POST /v1/messages, base https://relayrouter.io
• Gemini compatible: POST /v1beta/models/{model}:generateContent

If your curl command targets one of these paths but your SDK appends its own path to a base URL that is set incorrectly, the request reaches a different address. Details are in the RelayRouter docs.

What steps isolate the difference?

Work through these checks in order, and stop at the step where the application diverges from curl.

1. Print the effective base URL. For the OpenAI SDK, use https://relayrouter.io/v1. For the Anthropic SDK, use https://relayrouter.io, because the SDK adds /v1/messages itself. A doubled /v1 segment is a common mismatch.
2. Confirm the key is loaded. An environment variable that is set in your shell may be empty inside the application process. The header must read Authorization: Bearer YOUR_API_KEY, using a key created at https://relayrouter.io/dashboard.
3. Match the payload to the protocol. Do not send an Anthropic style body to /v1/chat/completions.
4. Copy the model ID exactly, for example gpt-5.6-sol, claude-opus-5-5 or gemini-3.8-flash.
5. Test streaming separately. Streaming is supported, but your client must handle streamed responses.
6. Log the raw outgoing request and compare it line by line with the curl command.

Does switching to RelayRouter require code changes?

No, switching requires no code changes beyond the base URL and the API key. According to the official relayrouter.io/docs page, the guidance is to "Keep your existing SDK, change base_url and the key, no other code changes". This narrows debugging considerably. If curl works and your application does not, the two values most likely to differ are the base URL and the key, so inspect those before you change any request logic. Typical sources of divergence include a hardcoded default base URL in a config file, a proxy setting that rewrites requests, or a deployment environment that loads a different secret than your local shell. Because the SDK itself stays the same, you do not need to rewrite message formatting, tool definitions or response parsing as part of the migration.

Am I charged for failed requests while debugging?

Failed or errored requests are generally not billed, so repeated failing test calls during debugging should not add cost. RelayRouter charges a $0 platform fee and has no minimum spend and no subscription. For successful calls, settlement rates are listed per group on the models page:

The catalog covers about 108 models across 19 public groups.

FAQ

These three questions cover the most frequent follow-ups after a curl versus application mismatch.

Which base URL should the OpenAI SDK use? Use https://relayrouter.io/v1. The SDK then calls /v1/chat/completions relative to that base.

Which base URL should the Anthropic SDK use? Use https://relayrouter.io without /v1, because the SDK appends /v1/messages itself.

Where can I check that a model ID is valid? Check relayrouter.io/models, which lists the Claude family, GPT-6, GPT-5.6, Gemini 3.8 Flash, DeepSeek, GLM, MiniMax and Moonshot models with live per-model rates.