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

# Workspace File Preview

> Click any file in the workspace browser — PDFs, images, audio, video — and preview it in-app with smart text/binary detection, streaming downloads, and safe headers.

# Workspace File Preview

The workspace file browser is where your Agent's sandbox files live. Click a file and Myrm decides how to show it — **text renders as text, and rich media renders in-app without leaving the workspace**.

## What previews in-app

| File type                     | Behavior                                                |
| ----------------------------- | ------------------------------------------------------- |
| PDF                           | Rendered inline with react-pdf, scrollable and zoomable |
| Image (PNG/JPEG/WebP/…)       | Full-size in-app render                                 |
| Video / Audio                 | In-app players over a streaming URL                     |
| SVG                           | Inline vector render                                    |
| Markdown / code / plain text  | Text viewer with the existing code/editing surface      |
| Unknown or unsupported binary | Download fallback — never garbled text                  |

## Smart text vs binary detection

Myrm does not guess. The backend marks every file entry with an authoritative `is_text` flag (zero-IO extension detection during the scan), and the frontend routes on that flag:

* **Text files** are served with a size cap so a huge log never freezes the page.
* **Binary files** stream in full over HTTP Range — the whole PDF, not a truncated fragment — so previews and downloads stay intact.

Unknown extensions are safe too: if the backend cannot classify a file as text, it is treated as binary and offered for download instead of rendering as mojibake.

## Safe downloads

Download responses protect against header injection (CR/LF stripped) and encode non-ASCII filenames per RFC 5987, so files with Chinese, Korean, or emoji names download correctly on any platform.

## Preview kinds

Artifact renderers are reused across the chat and the workspace browser: PDF, document, spreadsheet, image, video, audio, SVG, and interactive UI previews all share one rendering pipeline. The workspace path adds the binary streaming and safe-header behavior on top.

## Related

* [Workspace Auto-Organize](./workspace-organize)
* [File Checkpoints](./file-checkpoints)
* [Artifact Sharing](./artifact-sharing)