FAQ and troubleshooting

Which base URL should I use?

Use the protocol path for your client:

Client	Base URL
Claude Code	`https://api.fhrouter.com`
Codex CLI	`https://api.fhrouter.com/v1`
OpenAI-compatible apps	`https://api.fhrouter.com/v1`
Gemini CLI	`https://api.fhrouter.com/v1beta`
Antigravity	`https://api.fhrouter.com/v1beta`

Do I need any advanced settings?

No. Most users only need:

a token key
the correct base URL
the exact model id

Use the model ids shown in the console or returned by /v1/models.

Why is a request slow?

Large prompts, long output, high reasoning models, and model-side load can all affect latency. For first tests, use a small prompt and a small max_tokens value.

If a coding client appears to think for a long time and returns no visible text, stop the run, check FHRouter usage, then retry with a short prompt. Some clients stream or retry internally, so a partially processed request can still use quota.

How are tokens billed?

Requests are billed from model usage. Input tokens, output tokens, and reported usage are counted according to the active model pricing.

Check the usage page after a small test request before sending larger workloads.

Gemini gives an API key or model error

Common causes:

The client is using /v1 instead of /v1beta.
A local .env file is overriding Gemini CLI config.
The model id is not in the FHRouter model list.
The client is using OpenAI-compatible conversion instead of Gemini-native mode.

Use:

https://api.fhrouter.com/v1beta

If Gemini CLI runs in a sandboxed mode, make sure GEMINI_API_KEY and GOOGLE_GEMINI_BASE_URL are passed into the sandboxed environment.

A client passes its connection test but chat fails

Some clients enable tools, skills, or protocol conversion by default. Try:

turning off tools or skill calling
using a smaller prompt
selecting the correct API type
manually adding the exact FHRouter model id

Also make sure the app is not silently rewriting Gemini or Claude requests into OpenAI format.

I see `model_not_found`

The model name is wrong or not enabled for your account. Call:

curl https://api.fhrouter.com/v1/models \
  -H "Authorization: Bearer YOUR_FHROUTER_API_KEY"

Then use one of the returned ids.

I see `Selected model is at capacity`

This usually means the selected model is temporarily overloaded. Retry later or switch to another model id from the FHRouter catalog.

HTTP status guide

Status	What it usually means	What to try
`400`	Request format, unsupported beta header, or incompatible tool call	Use the matching protocol path, lower complexity, or disable experimental client features
`401`	API key rejected or wrong endpoint	Check the FHRouter token key and base URL
`404`	Wrong endpoint path or model id	Copy the model id from FHRouter and check `/v1`, `/v1beta`, or host root
`429`	Rate limit, cache limit, or too many retries	Start a new conversation, wait briefly, or reduce concurrency
`503`	Model temporarily unavailable	Retry later or switch model
`524`	Request timed out before the gateway received a final answer	Use a shorter prompt/output limit and retry

FHRouter hides detailed internal errors from normal user logs where possible. For deeper troubleshooting, note the approximate request time and status code before retrying.

The model list is empty

Check that you are signed in and that your token key has not been deleted. For API checks, call:

curl https://api.fhrouter.com/v1/models \
  -H "Authorization: Bearer YOUR_FHROUTER_API_KEY"

If the API returns models but the dashboard does not, refresh the page and sign in again.

Payment succeeded but balance did not update

Wait for the browser to return to FHRouter after payment. If the payment page was closed early, refresh the billing page after a short delay.

If it still looks wrong, note:

FHRouter account email
payment time
payment amount
screenshot of the payment status

Those details are usually enough to trace the payment in the billing records.