A chatbot is only as good as the source it draws on. EdTechRAG is filled with your own content and answers questions from it — not from a language model’s general knowledge. The source code is open and the application can be self-hosted.
The method behind it is called retrieval-augmented generation (RAG): for a given question, the matching passages in your own documents are located first, and only those passages are passed to the language model. The answer therefore stays tied to the stored content and can be traced back to its source.
Metadata instead of a wall of text
Plain similarity search over chunks of text often finds the right wording but the wrong document. EdTechRAG therefore carries metadata alongside the content — narrowing the search before it starts. What was “it said that somewhere” becomes a specific hit that teaching staff and students can go and read for themselves.
Open and self-hostable
The code is available on GitHub under the MIT licence. Self-hosting requires:
DockerandNode.js- PostgreSQL with the
pgvectorextension for vector search - optionally a GitHub or GitLab connection so that content is ingested automatically
Anyone self-hosting EdTechRAG must remove all references to TU Graz — logo and imprint — unless this has been expressly agreed. The copyright notices and the licence attribution have to be kept.