> ## Documentation Index
> Fetch the complete documentation index at: https://docs.depost.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Attachments and voice input

> Bring a file, a library document, or a spoken note into a chat.

Context sources give chat access to your material in general. Attachments do something
narrower: they put one specific thing in front of it for this request.

## Attach a file

Click **+** next to the message box and choose **Add photos & files**.

<Frame>
  <img src="https://statics.depost.ai/docs/chat/attachments-add-menu.png?v=2" alt="The plus menu open beside the chat message box, showing Add photos and files, and Select from library" />
</Frame>

Supported formats:

| Type         | Formats                 |
| ------------ | ----------------------- |
| Documents    | PDF, DOC, DOCX, TXT, MD |
| Slides       | PPTX                    |
| Spreadsheets | XLS, XLSX, CSV          |
| Data         | JSON                    |
| Images       | PNG, JPG, JPEG          |

Attach a file when the request is *about that file* — turn this transcript into three
posts, pull the findings out of this report, write about the case study in this deck.

## Pull in a library document

The same **+** menu has **Select from library**, which opens your content library with a
search box. Use this instead of re-uploading something you've already saved.

<Tip>
  Attaching one library document is more precise than switching on the whole **Content
  Library** context source. Use the attachment when you know exactly which document
  matters, and the context source when you want Depost AI to find what's relevant.
</Tip>

Files that have been brought into a conversation are collected under **Files in chat**, so
you can see what the thread is working from.

## Voice input

Dictate instead of typing. The microphone sits at the right of the message box, next to
send.

<Frame>
  <img src="https://statics.depost.ai/docs/chat/attachments-voice-button.png?v=2" alt="The message box with a Voice input tooltip showing above the microphone button" />
</Frame>

While you talk, the box becomes a waveform with a running timer. The cross discards the
recording. The tick ends it and transcribes what you said.

<Frame>
  <img src="https://statics.depost.ai/docs/chat/attachments-voice-recording.png?v=2" alt="The message box recording, showing a waveform, a timer reading fourteen seconds, and buttons to discard or accept" />
</Frame>

The transcription arrives in the message box as editable text, so nothing is sent until you
have read it.

This is genuinely useful for a specific case: describing an idea out loud usually produces
more detail than typing it. If your prompts tend to be one line, try saying the same thing
instead — you'll give Depost AI noticeably more to work with.

<Steps>
  <Step title="Press the microphone and talk it through">
    Say who the post is for, what happened, and what you want the reader to do. Don't
    script it.
  </Step>

  <Step title="Stop when you're done, and read it back">
    The tick transcribes. Fix names and technical terms, which are what dictation gets
    wrong.
  </Step>

  <Step title="Send it as the brief">
    A rambling spoken brief beats a tidy one-line prompt almost every time.
  </Step>
</Steps>

## Related

<CardGroup cols={2}>
  <Card title="Give chat the right context" icon="layers" href="/chat/context">
    Choose which of your material Depost AI draws on.
  </Card>

  <Card title="Library" icon="folder-open" href="/library/overview">
    Save documents once and reuse them across chats.
  </Card>
</CardGroup>
