Retrieval-augmented generation (RAG)
A technique that retrieves relevant source material and supplies it to a model as context before generation, so the output is grounded in specific documents rather than the model's parametric memory.
In more detail
Retrieval-augmented generation addresses a basic limitation: a model only knows what it was trained on, and it can state things confidently that are wrong or out of date. RAG inserts a retrieval step first. The system finds material relevant to the question, usually via embeddings, and passes that into the model's context so the answer draws on real, current sources.
RAG improves accuracy and lets a model work with private or recent information it never saw in training. It also creates a natural place to attach citations, since the retrieved passages are known. The retrieval quality is decisive: if the wrong passages are retrieved, the generation is grounded in the wrong thing.
Where this shows up at Ceven
Ceven's wide and deep research follow the RAG pattern at their core: gather and rank sources, then generate a brief grounded in them with citations, rather than emitting an unsupported summary. The same grounding pattern lets workflow steps answer from the customer's own documents instead of relying on model memory alone.