Title: How the AI help works · Help · Steerbay
URL: https://www.steerbay.com/en/docs/ai-guide/
Language: English (Česky: https://www.steerbay.com/cs/docs/ai-guide/)
Summary: Answers about the platform from the public guides, with links to the sources.

---

# How the AI help works

Answers about the platform from the public guides, with links to the sources.

## An answer instead of searching every guide

The help answers questions about the platform from these public guides and links the sections it relies on. It answers only questions about the platform and politely declines other requests, such as writing an essay or code. Answers can contain errors; open the source for important decisions. When the guides do not contain an answer, the help says so and suggests the closest guide.

## Connecting the model

AI answers work when the server has an Anthropic API key configured. Until it does, the website says so clearly and the guides can still be read and searched; no pretend answers are produced. The help does not carry out browser tasks, manage a team or activate an account. For work with the agent, open the app.

## Public questions, not company material

Your question, a short history of this conversation and the public guides are sent through the server to the Anthropic API. The service is not meant for personal data, passwords, invoices or confidential documents. The conversation is not stored on a server or in a database. It stays only in this browser, so you can return to it after closing the panel or on another page; New conversation deletes it, and it disappears by itself after 30 days without use. This is not a statement about Anthropic's or the hosting provider's data-retention policies.

## How many questions you can ask

To keep the help available to everyone, questions are limited: at most 5 a minute, 30 an hour and 80 a day from one connection, plus a shared daily limit for the whole website. When you reach a limit, the help says how long until you can ask again; the guides stay available. Counting uses only a one-way fingerprint of the connection's address, never the question, and it is deleted within a day. Repeatedly sending messages that are not questions (spam, gibberish, attempts to get around the rules) switches the help off for that connection for an hour, and for a day if it happens again. Before each question your browser does a short check that it is not an automated sender; you will rarely notice it.

Missing an answer? The AI help searches every guide and shows its source.