# FAQ

https://modelportal.ai/docs/faq

## Setup

**What base URL do I use?**
OpenAI-compatible tools take `https://modelportal.ai/v1`; Anthropic-compatible tools such as Claude Code take `https://modelportal.ai` without `/v1`; for Gemini see the [Gemini API](/docs/gemini-api). Per-tool settings are in the [Integration guides](/docs/integrations).

**Why is the model list empty?**
`GET /v1/models` returns only models this key may call that have an available endpoint. Check the key's allowed models, or type the model ID in by hand.

**Can I use my key directly in a browser or app?**
No. A key in frontend code can be pulled out by anyone and spent against your balance. Route calls through your own backend.

## Billing

**Am I charged for failed calls?**
No. Failed, canceled and unknown-outcome calls are free; see [Attempts and settlement](/docs/attempts).

**Why do I get 402 when my balance looks sufficient?**
Before a call the platform holds the input estimate plus the allowed maximum output. Lower `max_tokens`, or top up and retry.

## Routing

**Will the platform swap my request to another model?**
No. The model you request is the model that runs; the platform only chooses among that model's endpoints.

**How do I choose a provider?**
Add a `provider` object to the request, for example `{"order": ["PROVIDER_ID"]}`; see [Provider selection](/docs/provider-routing).

**How do I keep a multi-turn conversation on one endpoint?**
Send the same `X-ModelPortal-Session` header with every request; see [Default routing](/docs/default-routing).

**Does a failed first endpoint fall back to another provider?**
Not by default: `allow_fallbacks` defaults to `false`. Turn it on and errors such as connection failures, timeouts, 429 and 5xx move the request to the next endpoint of the same model, up to 3 endpoints. See [Fallbacks](/docs/fallbacks).