# Chat with your files

The **Chat** screen lets you ask questions about your files in plain language. The model
searches your library, reads the relevant files, and answers using what it found. It
shows its sources as you go.

{/* SCREENSHOT: a chat answer with two citation cards underneath, one being hovered */}

## How a chat works

Type a question and send it. The model searches your files, reads the relevant ones, and
writes a response based on their content. The files it used appear as clickable citation
cards. Click a card to open that file in **All Files** and verify the source.

## Start fresh or pick up an old thread

- **New chat** clears the conversation and starts over.
- **Past conversations** holds your history so you can reopen an earlier thread and keep
  going.

## Chat across every source

By default, Chat answers from the current storage source. Switch the **Storage Source**
selector to **All Sources** to let Chat draw on every source at once. See
[Manage storage sources](/using/manage-sources/).

## Ask better questions

- Be specific. "What were the risks flagged in the 2024 vendor contracts" beats "tell me
  about contracts."
- Name what you are after. A date range, a project name, a file type, or a person narrows
  the results.
- Ask follow-ups. The conversation keeps context, so you can refine without restating
  everything.

> [!NOTE]
> Chat needs a **Chat** model assigned in **Settings**, and it answers best when your files
> have been processed. See [Models](/reference/models/).