Using the AI prompt workspace

What the in-product prompt assistant can do, how sessions are stored, and why exports are part of the chat workflow.

Client users and analysts 7 min read

Prompting is document-specific, not a generic chatbot

The default assistant message explicitly asks questions about business documents and gives export examples such as showing invoices and exporting to CSV.

This is a retrieval-and-action surface tied to document data, not a marketing assistant or general support bot.

Sessions and history matter

The UI can create new chats, switch active sessions, fetch prior messages, and reopen session history. The backend stores titles derived from the user prompt, which means recurring analytical conversations are expected.

The assistant can return files

The prompt page has dedicated logic for `downloadUrl`, `fileName`, and downloadable message variants. Knowledge content should teach users that some prompts return datasets, not only plain text responses.

Credit awareness is built into the experience

The UI logs token usage metadata in the browser console per action and model. Even if end users never inspect that, it is a sign that prompt usage has operational cost and should be documented with concrete examples.

Related

Continue with adjacent workflows