FAQ

These are the most common questions during setup. When troubleshooting, confirm three things first: the Base URL is correct, the API Key is valid, and the model name matches the console exactly.

What should the Base URL be?

OpenAI-compatible clients usually use:

https://app.yylx.io/v1

For Anthropic-protocol tools like Claude Code, use the Claude/Anthropic address shown in the console.

Don't put a full endpoint path into the Base URL. OpenAI-compatible clients generally shouldn't include paths like /chat/completions or /responses — the client appends those itself.

Can multiple tools share one API Key?

Yes, but it's not recommended. Separate Keys per tool make usage tracking and leak handling much easier.

Recommended per-tool naming:

Can I view the full Key again after creating it?

Don't rely on being able to. Copy and save the Key immediately after creation. If you forget, the safest fix is to delete the old Key and create a new one.

Why isn't my model name accepted?

Make sure the model name matches the console exactly. Model names are usually case-sensitive, and stray spaces break them too.

If your client supports adding models manually, copy the full model name straight from the console. Don't abbreviate names that carry dates, suffixes, or hyphens.

Why does the client report 401?

Usually the API Key is invalid, incompletely copied, padded with whitespace, or already deleted. Copy it again or create a new one.

If you just disabled, deleted, or recreated a Key, make sure the client has been updated to the new Key. Some desktop clients cache old configurations and need a restart after saving.

Why does the client report 404?

Most often the Base URL is wrong, or a full endpoint path was pasted into it. OpenAI-compatible clients generally only need the address up to /v1.

Claude Code can also fail with https://app.yylx.io/v1, because it uses the Claude/Anthropic protocol configuration. Go back to the API Keys page, click "Use Key", and copy the Claude Code configuration.

Why does the client say the model or API format is unsupported?

That's usually a protocol mismatch. Codex, Cursor, and Cherry Studio mostly use the OpenAI-compatible API, while Claude Code uses the Claude/Anthropic environment variables. Make sure you haven't put the Claude Code address into an OpenAI client, or the OpenAI /v1 address into Claude Code.

Why are responses slow?

It can come from the model itself, the upstream route, concurrency, or your network. Try a different model, and check the console announcements or the status page.

Test with a short prompt first — for example, ask the model to reply with a single sentence. If short prompts are fine but long tasks are slow, it's usually model inference or context length; if short prompts are also slow, look at the network, service status, or the current model's availability.

How do I confirm requests actually go through yylx.io?

After setup, send a short message, then check the usage log in the yylx.io console. If a request shows up at the matching time, the client is going through yylx.io. If not, the client is usually still using an old provider, stale environment variables, or its built-in service.

What if my Key leaks?

Immediately disable or delete the leaked Key in the console, then create a new one and update the client configuration. If you're unsure about the scope of the leak, also review the recent usage log for unexpected consumption.