What is a Large Language Model (LLM)?
What is it?
A Large Language Model (LLM) is a neural network trained on vast amounts of text to predict and generate language. You give it a prompt (input text), and it produces a completion (output text) — a sentence, paragraph, or full answer.
Examples include OpenAI's GPT family, which Noulek uses for generation.
Why does it exist?
LLMs can read natural language, follow instructions, summarize, translate, and answer questions in a conversational style. They replace rigid rule-based systems for many language tasks.
In production systems, LLMs are rarely used alone for factual business answers — they are combined with retrieval (RAG) so responses can cite real documents.
How does it work conceptually?
At a high level:
- Text is split into tokens (word pieces).
- The model processes tokens through many layers of computation.
- It predicts the next token repeatedly until the answer is complete.
You do not need to understand the math to use an LLM. The important idea: the model generates text based on patterns learned during training plus whatever you put in the prompt at request time.
Example
Prompt: "Summarize the following maintenance note in one sentence: …"
The LLM reads the prompt and writes a summary. If the prompt includes retrieved manual text, the summary can reflect that specific content.
Where does it fit in a RAG pipeline?
In RAG, the LLM is the generation stage — the last step:
Question
↓
Retrieval (find relevant chunks)
↓
Context assembly (chunks + question in prompt)
↓
LLM (generate answer)
↓
Answer (ideally grounded in context)Search without RAG may stop at retrieval and show chunks. Chat with RAG sends retrieved chunks to the LLM to produce a natural-language answer.
How this relates to Noulek
Noulek configures the LLM in Settings → LLM:
- OpenAI API Key — credentials for the provider
- Model — which model tier generates answers (Luna, Terra, Sol in the Noulek UI)
The same API key is also used for embeddings during document indexing. Chat and RAG answers require a working LLM connection. Use Test connection to verify setup.
If the LLM fails, you may see errors like LLM error or connection test failures. See LLM connection problems.
Noulek does not train its own LLM — it calls a configured provider (OpenAI) with retrieved context from your Knowledge Base.