# FAQ

Ordered by expected frequency, not by topic.

### How is this different from calling an LLM directly?

Answers are retrieved from a scholarly corpus and every claim carries a checkable source with a page
locator. Claims the model could not ground are labeled. See citation.

### What is the difference between a paper and a document?

A paper is in our corpus; a document is yours. See paper and document.

### When should I use search vs chat?

Search finds papers and answers across the corpus. Chat answers questions about a fixed set of
sources you name. See search.

### How much does a call cost?

Metered in credits, reported per response in `X-Credits-Cost`. See pricing-and-credits.

### How long does a deep review take?

Minutes. Stream it or use a webhook. See streaming-deep-review.

### Can I store the results?

Depends on the content type and licence. See content-licensing.

### Do you train on my uploads?

See data-privacy.

### Can I call this from the browser?

Not with a secret key. Proxy through your backend. See authentication.

### Which languages are supported?

Input in 75+ languages; generated output in 24, including English, Spanish, Portuguese, French, German,
Chinese, Japanese, Korean, Arabic, Hindi, and Bengali. An unsupported `language` returns
`422 language_unsupported`.

### Do you have an OpenAPI spec?

See reference-overview.

### Is there a free tier?

Yes. The Developer plan is free with 500 credits a month — 100 standard searches — plus 1,000 one-off
credits during public beta. Test keys have their own free 500/month. See
pricing-and-credits.