Adding your sources

Sources are the material a course is built from: your policies, procedures, slides, recordings. You give them to Vidext's agents — the AI agents that plan and write your courses — in the chat, and every file you add becomes one of your organization's documents. This page covers what you can add, how, and what Vidext does with it.

Attach filesCopy Anchor Link

In the chat's message box:

• Select the plus button (Add attachments), then Attach files or images.
• Or drag files onto the chat. It shows Drop files here.
• Or paste a picture from your clipboard. It becomes a file, not text.

Each file appears as a tile above the message box, with its progress. You can send your message while files are still processing: the conversation shows File ready when the agents can read a file, and they pick it up from there. You can also attach and send while the agents are answering another message: yours waits in Up next and is read once they finish. A file sent in a module's own conversation, the agent's tab beside the studio, is that module's: File ready shows in that tab, and the module's agent picks it up even if it already answered your message. To drop a file before you send, remove its tile.

You can attach up to 20 files per message.

Add a YouTube videoCopy Anchor Link

Paste a YouTube address on its own in the message box, or drag the link in from a browser tab (Drop a YouTube link here). It becomes a tile like a file. A YouTube address inside a sentence stays part of your message.

Watch, Shorts, embed, live and youtu.be links all work. Vidext uses the video's captions when YouTube publishes them, and in every case watches the video, so what it shows is kept next to what it says. A video without captions is still added. Playlists, channels, and private or members-only videos are not supported.

Only YouTube links can be attached. Any other link gets Only YouTube video links can be attached.

Share a web pageCopy Anchor Link

Type the page's address in your message. The agents read it and tell you what the page covers. If the page cannot be read, for example because it needs a sign-in, they say so and ask for it as a file. For pages behind a sign-in, such as Google Docs or an intranet, download the file and attach it.

Supported files and limitsCopy Anchor Link

Older Office formats (.doc, .ppt, .xls) are not accepted: save the file as .docx, .pptx or .xlsx first. Code files are read as plain text. A WebM file is sorted by what it holds: with a video track it is a video, with sound only it is audio.

When a file cannot be added, its tile turns red and says why:

What Vidext does with a fileCopy Anchor Link

Each file is processed in the background:

• Text is extracted, and usable images are pulled out of documents.
• Text in pictures is read. A screenshot, a slide or a photo of a page in PNG, JPEG or WebP goes through text recognition, so what it says counts like a document. A GIF is described instead. A scanned PDF, or the pages of a PDF that have no text of their own, are read the same way.
• Audio is transcribed.
• Video is transcribed with each speaker labelled (Speaker 1, Speaker 2: the labels tell voices apart, they do not identify anyone) and summarized. Later the agents can read what one speaker said or what was said at a given moment, and they watch the video when you ask about something that is only shown on screen. A video without usable speech is watched and described.
• Linked and embedded videos are watched too. A YouTube link inside a document you upload, or a recording embedded in a PowerPoint or Word file, is watched, and what it says and shows is added to that document's text.
• The same file is not processed twice. Uploading a file the organization already has, or a video it has already watched, reuses the earlier result.

Your organization's documentsCopy Anchor Link

A file you attach belongs to the organization, not to one conversation or one course. The agents can read it in any later conversation, for any course. They also read the pages of your company website that Vidext read when your organization was created. See What Vidext reads.

Once the agents have used or read something in a conversation, a Sources button appears in its header and lists what they used and read there: files attached in that conversation, organization documents they opened, and web pages their research cites. See Working in the chat.

Web researchCopy Anchor Link

While they write the course brief — the short summary of who the course is for and what they will learn, which you approve before anything is built — the agents also research the topic on the public web and in research papers, to cover what your material leaves out. Pages they cite appear in the conversation's Sources. Research never stands in for your organization's own facts, such as a policy, a figure or a contact: when the course needs one of those and no source has it, the agents ask you for the document. If you want the course built only from your material, say so and they stop researching.

Services you connect in Settings → Connectors, such as Google Drive or Notion, are not read by the agents yet. Download the file and attach it instead.

Choosing good sourcesCopy Anchor Link

• Prefer the source of truth. The actual policy or procedure beats a slide summary of it.
• Fewer, better documents. Five precise documents do more than fifty loosely related ones.
• Recordings work. A recorded walkthrough or an expert explaining a process is a real source.
• No document? Describe what you know in the conversation. The agents ask questions while they plan the course, and ask for the specific documents the course needs. See Plan a course.