DGTLMS & LMSDISPATCHER Guide

Artificial intelligence

Index content for the assistant (RAG)

Index your courses so that dgtTutor answers from the actual content — this is retrieval-augmented generation (RAG).

Administrator~5 minutesIntermediate
  • A configured embeddings backend: a Mistral / OpenAI / Gemini key, or a local server (Ollama).

How it works

Indexing extracts the text from the content (HTML, SCORM/Storyline, PDF, subtitles…), splits it into chunks and vectorises it. dgtTutor then relies on the most relevant extracts to answer.

Steps

  1. Global indexing

    Administration → Course management → Run learning-path indexing (AI). Progress is shown live, with the Learning paths / Indexed / Unchanged / Chunks / Skipped (dispatched) counters. The operation is resumable: running it again only reprocesses modified content.

  2. Or per course

    The Index content for AI button in a course's editor.

  3. Run it again after an import

    Re-index after adding or modifying content.

Extraction covers HTML, DGT Studio and SCORM / Storyline packages. Without a configured embeddings backend (Mistral / OpenAI / Gemini key, or local endpoint), the button stays disabled and a message says so.

Dispatched content (hosted by the publisher)

Some SCORM packages (e-doceo, Talentsoft, dispatched Rise…) only contain a player shell: their text is served by the publisher's server. Global indexing counts them under Skipped (dispatched); a second indexing run downloads their content (https, public servers only) and then indexes it.

  1. Open the block

    On the same page, Dispatched (remote) content block.

  2. Run it now

    Index dispatched content now button; progress is shown below it.

  3. Or every night

    Leave the Automatically index dispatched content every night box ticked (ticked by default). The task resumes where it stopped.

The block shows the result of the last run, manual or nightly, with an example error: content protected by authentication, publisher's server unavailable… Unchanged content is not re-vectorised.

  • Vectorisation can be 100% local (Ollama): no content leaves the server, at zero cost per call.

See also