# Images & files

> What your AI Agent perceives when a visitor sends an image or a file. Images are analyzed visually; files are recognized by name, and their contents are not read.

When a visitor sends an attachment during a conversation, your AI Agent treats **images** and **other files** differently. Knowing the difference helps you set the right expectations and write better Guidelines.

## Images

When a visitor sends a **photo or image**, your AI Agent actually **sees and analyzes the image content**. It can describe what's in the picture, read text visible in it, and answer questions about it — the image is sent to the underlying vision-capable model alongside the visitor's message.

This works on every plan, and on every channel where a visitor can attach an image (Telegram, WhatsApp, Intercom, HubSpot, and images supplied through the [Quickchat API](/channels/api/)).

:::note
If an image can't be loaded (for example, an expired or unreachable link), the AI Agent simply continues with the text of the message instead of failing the reply.
:::

## Files (PDFs, documents, and other attachments)

When a visitor sends a **non-image file** — a PDF, a Word document, a spreadsheet, an audio or video file — your AI Agent is **told that a file was sent and its filename**, but it **does not read the file's contents**. The Agent knows, for instance, that `invoice_2026.pdf` arrived, so it can acknowledge the file and respond sensibly, but it cannot extract or summarize what's inside it.

Inbound file awareness is available on these channels:

| Channel | Visitor sends an **image** | Visitor sends a **file** (PDF, doc, ...) |
|---|---|---|
| [Telegram](/channels/telegram/) | Analyzed visually | Filename only |
| [WhatsApp](/channels/whatsapp/) | Analyzed visually | Filename only |
| [Intercom](/channels/intercom/) | Analyzed visually | Filename only |
| [HubSpot](/channels/hubspot/) | Analyzed visually | Filename only |

:::tip
If you want the AI Agent to help with file contents, ask the visitor to paste the relevant text into the chat, or rely on **Human Handoff**. The default **Media Message** handoff rule escalates a conversation to a human when a visitor sends an image or file the AI can't process — see [Actions → Handoff Rules](/ai-agent/actions/#handoff-rules).
:::

:::caution
This is separate from the **Knowledge Base**, which is text-only: you can upload PDFs and documents there and Quickchat extracts their text for the Agent to use as knowledge (see [Upload File](/ai-agent/knowledge-base/#upload-file)). The behavior on this page is about files a **visitor** sends mid-conversation, which are not added to the Knowledge Base.
:::