Skip to content

Professional services

Need expert help? RAGcon GmbH offers:

  • RAGflow Support
  • RAGflow Hosting
  • Moodle Support
Talk to RAGcon GmbH →

FAQ

Answers to recurring questions about the Moodle RAGflow suite. Missing something? Open an issue on the relevant plugin's repository (linked at the top of each plugin page), or see Help & Support.

Answers, sources and language

Why does the Tutor sometimes say the answer isn't in the course materials?

That is intended. The Tutor answers only from the knowledge base for that course and will not invent a plausible-sounding answer when the materials don't cover the question — for a course tutor, declining is safer than guessing. If something should be answerable, add or re-process the relevant document (see Tutor block), or rephrase the question. A greeting or small-talk can trigger a similar line — see the next entry.

The assistant greets the user but adds “The answer you are looking for is not found in the dataset!” — even to a plain “hello”. Why?

That sentence comes from the RAGflow assistant's own system prompt, not from the plugin. RAGflow's default assistant prompt contains a fixed instruction along the lines of: “When all dataset content is irrelevant to the question, your answer must include the sentence 'The answer you are looking for is not found in the dataset!'.” A greeting matches nothing in the knowledge base, so the model both greets and appends that mandated sentence.

This only happens when the block is pointed at an assistant that was created in RAGflow's own web UI (which seeds that default prompt). Knowledge bases/assistants created from the block get a clean prompt that does not force the sentence. The block's own System instruction setting cannot override it either: RAGflow's chat endpoint ignores a per-request system message — the assistant's stored prompt always wins.

Fix it one of two ways:

  • Recommended — use a block-created assistant. In the block's Knowledge base / assistant setting choose ➕ Create new knowledge base … and add the course documents there. The new assistant is given the clean prompt automatically.
  • Or edit the existing assistant in RAGflow. Open the assistant in RAGflow → Prompt and remove the “…must include the sentence …not found in the dataset!” clause (and clear Empty response if it is set). RAGflow only applies such prompt changes in its own UI; the plugin cannot rewrite an existing assistant's prompt through the API, which is why a block-created assistant is the simpler route.

See Tutor block → Knowledge base / assistant for where the assistant is chosen.

The assistant greets the user but lists sources (e.g. [1.1] [1.2]) that have nothing to do with a plain “hello”. Why?

Same cause as the entry above. RAGflow's default assistant prompt tells the model to summarise and cite the dataset on every turn, so it attaches citation markers even to a greeting. The plugin does not add sources of its own — it turns the model's own citation markers into the Sources list — so a spurious citation becomes a spurious source. It is inconsistent (the model only sometimes adds the citations), which is why the same greeting shows sources one time and none the next.

With a clean prompt the model only cites a document it actually used, and a greeting shows no sources. Fix it the same two ways: prefer a block-created assistant (clean prompt automatically), or edit the existing assistant's prompt in RAGflow and drop the “summarise / list the dataset” instruction. See the entry above and Answer wording when nothing is found.

Can I trust the sources listed under an answer?

The citations come from the documents the model actually used to write the answer (its own reference markers), not a separate keyword search — so the list reflects the real basis of the answer. (The one exception is an assistant left on RAGflow's default prompt, which can cite documents even for a greeting — see the entries above.) As with any AI, open the linked source to confirm the detail that matters. Retrieval-only results in the Search block are even more direct: every hit is a real document you can open.

In which language does it answer?

In the user's Moodle language, taken from their profile or the course's forced language — independently of the language your documents are written in. You don't configure this per plugin.

Is my chat private — can teachers or admins read it?

Conversations are per user. The Helpdesk has a private/incognito mode and a forget memory control, and long-term memory is opt-in. The usage log stores metrics only — no message content. See Security and data protection.

Documents and understanding

What does “deep document understanding” mean?

RAGflow reads the content of your files, not just their titles: OCR turns scanned PDFs and photographed pages into text, layout recognition captures titles, tables, figures and reading order, and an optional vision model describes images and diagrams so their content becomes searchable. On RAGflow 0.25+ MinerU and Docling are additional parsing methods (formula and table recognition). The result: scans, images and tables become answerable, not just plain text. See Document understanding.

Which file types can it read — and does it handle scans and images?

Common document formats (PDF, Word, PowerPoint, Excel, HTML, Markdown, plain text), plus scanned PDFs and images via OCR and — for the content of pictures and diagrams — an optional vision model. Whether image understanding is available depends on the RAGflow models your administrator has configured; see Set up RAGflow or ask your Moodle/RAGflow administrator.

I uploaded a document but the Tutor doesn't know it yet — why?

RAGflow parses each document first, which takes anywhere from seconds to minutes for a large scanned file. The knowledge-base panel shows a status indicator; the document is available once it turns green. See Tutor block.

How do I keep the knowledge bases up to date?

Either manage documents in the block (upload / re-process / delete), or use the RAGflow Moodle Connector to sync a Moodle course's content into the knowledge base automatically, so answers reflect the current materials without manual re-upload.

Tutor, Search and Helpdesk

Why does RAGflow find more than Moodle's built-in search?

Moodle's search is built to find courses and activities by name — it matches keywords against titles, text fields, metadata and forum posts. It does not look inside your files.

RAGflow reads the content of the documents themselves. Its DeepDoc engine applies OCR (so scanned PDFs and photographed pages become text), recognises layout (titles, tables, figures, equations and reading order) and captures table structure; an optional vision model even describes images and diagrams, so their content becomes searchable and can appear as a source. Retrieval then matches by meaning (embeddings, optional rerank), not exact keywords, and returns the supporting passage and image — which the Tutor and Helpdesk can turn into a cited answer.

In short: Moodle's search finds where something is; RAGflow finds what a document says. See Document understanding for the details.

What's the difference between the Tutor and the Search block?

The Tutor answers questions in a chat, grounded in a course's knowledge base and with citations. The Search block only finds documents — a ranked list of sources with the matching passage, no generated answer. Use Search when you want the file; use the Tutor when you want an answer.

Tutor or Helpdesk — which one do I use?

The Tutor is per course, grounded in that course's materials, and lives on the course page. The Helpdesk is site-wide, answers from an organisation-wide knowledge base, and is reached from the site's user menu on every page.

The Tutor or Helpdesk says “no information” even though the knowledge base has the answer — why?

Almost always the RAGflow assistant's retrieval settings are too strict for that knowledge base. The usual culprit is a rerank model on the assistant: its scores sit on a lower scale, and the assistant's Similarity threshold is compared against them, so on a small or short-document knowledge base every passage falls below the cut-off and the assistant answers “no information”. Fix it in the RAGflow assistant — lower the Similarity threshold (around 0.1) or remove the rerank model. See Set up RAGflow.

The Search block returns nothing or very few results — is it broken?

Probably not. A relevance floor hides weak matches on purpose, so “no strong match” is a valid, honest result rather than a padded list. Try rephrasing or broadening the query; an administrator can tune the threshold and an optional rerank model per block — see Search block.

Setup and operation

If I point the provider's generate_text action at a RAGflow assistant, does that affect Moodle's built-in AI (editor, course assistant)?

Yes. Routing in Moodle's AI subsystem is by action, not by plugin: every generate_text / summarise_text / explain_text request that Moodle sends to the RAGflow provider — including Moodle's own placements (editor AI, course assistant) and any third-party AI placement — is answered by the assistant you configured for that action. If you don't want that, control it in the AI subsystem: disable the action on the provider instance (or don't make the provider available for that action), or turn the placement off. See AI provider and Security and data protection.

Do the Tutor and Search blocks go through Moodle's AI subsystem?

No. They use the provider's connection (base URL and API key) directly, with their own per-block knowledge base and assistant — not the core generate_text action. So a block's assistant/KB is independent of the provider's per-action configuration; changing one does not change the other. See AI provider.

Which RAGflow models do I need?

A chat model and an embedding model are required; an img2txt (vision) model is needed for image understanding, and a rerank model is optional. Your administrator sets these in RAGflow — see Set up RAGflow and the RAGflow documentation.

What data leaves Moodle, and where is it stored?

The suite talks only to your RAGflow instance — one you run yourself, or one RAGcon hosts for you — never a shared third-party service. Documents, conversations and embeddings live in that RAGflow instance; the provider's usage log in Moodle holds metrics only (no message content). Either way the data stays under your (or your hosting provider's) control — see Security and data protection, or ask RAGcon support about hosting.

Do I need the RAGflow Moodle Connector?

No — it's optional. Without it you manage documents directly in the blocks. With it, a Moodle course's content is synced into the RAGflow knowledge base automatically, which is useful when the material changes often. See the RAGflow Moodle Connector.

Is the usage dashboard required?

No. The dashboard is an optional add-on; the other four plugins (provider, tutor, search, helpdesk) work fully without it.

Note: This documentation was created with the help of AI. Spotted an error, an omission, or something unclear? Please report it in the ragcon-docs GitHub repository so we can fix it.