Wiki

Files and Attachments

You can attach files to your chat messages so DMJBot can work with them, and the bot can create files for you and return them in its reply. Everything happens right in the chat — there is nothing to configure.

Attaching files

In the chat input, pick one or more files and send them with your message. You can attach several files to a single message. Once sent, the files are available to the bot while it answers, so you can ask it to read, summarise, search, compare, convert, or analyse them.

What DMJBot can do with your files

Read text files. Plain-text files — .txt, .md, .csv, .json, source code, and similar — are read directly.

Extract text from documents. For binary documents the bot can pull the text out and work with it. Built-in formats:

• PDF (.pdf)
• Word (.docx)
• Excel (.xlsx)

More formats can be supported when a file-conversion tool is connected to your bot. After extraction the text is also saved as a companion file in the chat so it can be reused.

Analyse images. The bot can look at an image (.png, .jpg/.jpeg) — describe it, read text from it, or answer questions about it.

Images need a multimodal model. Image analysis works only when the model you have configured supports images (is multimodal). With a text-only model the bot cannot see image contents. See the Settings Guide for choosing a model.

Getting files back from DMJBot

You can ask the bot to create a file and it will return it in the response for you to download. The bot writes the content as text or Markdown and can convert it into a document format:

• PDF (.pdf)
• Word (.docx)

For example: "Write a one-page project summary and give it to me as a PDF" or "Turn these notes into a Word document." The conversion source is Markdown or plain text, and the resulting PDF/DOCX is attached to the reply.

Moving files between tools and devices

DMJBot can move a file from one tool, server or device to another in a single request — you don't have to download it and re-upload it yourself. If you have, say, the Gmail tool and a File Storage tool on your laptop connected, you can ask:

"Get the attachment from that email and save it to the Projects folder on my laptop storage."

The bot fetches the attachment from Gmail and writes it to the File Storage folder. The same works across any file-capable tools — email, File Storage, Dropbox, rclone remotes, Slack and Jira — between devices, in either direction. A few more examples:

"Download the latest report from my Dropbox Reports folder and upload it to the backups bucket on my rclone remote."

"Take the spreadsheet attached to that email and post it as a file in the Slack #finance channel."

"Grab the file from Projects/specs.pdf on my laptop storage and attach it to Jira issue ENG-512 as a comment attachment."

"Save the file from this Slack message to my Dropbox Inbox folder."

"Move every PDF in my Dropbox Scans folder to the Documents folder on my home server."

(Slack and Jira accept files too — Slack as a file posted to a channel or DM, Jira as an attachment on an issue or comment.)

This is optimized to use very few LLM tokens, no matter how large the file is. The file's actual contents never pass through the language model. When one tool produces a file, the bot keeps the data aside and hands the model only a tiny reference to it; when the next tool needs the file, that reference is resolved back to the real contents behind the scenes. So copying a 1 KB note and copying a 200 MB archive cost the model essentially the same — you are not billed tokens for the file itself.

Supported formats at a glance

Tips

• Keep filenames simple — letters, numbers and dashes — for the smoothest handling.
• Tell the bot what you want done with a file; it only reads or analyses contents when the task needs it.
• If image analysis doesn't work, check that your configured model is multimodal.
• Very large files take longer to process — attach only what is needed for the task.

See also

• Tools and MCP — file tools on your devices (File Storage, Dropbox, rclone).
• Settings Guide — choosing a model, including multimodal models.