Knowledge & Retrieval

RAG

Ground AI answers in your own documents

Retrieval Text Advanced Growing
All AI skills
Capability type
Retrieval
Modality
Text
Typical input
Question + document corpus
Typical output
Cited answer
Measured by
Groundedness / recall@k

Overview

RAG combines search with generation. Rather than trusting what a model memorised during training, the system retrieves relevant passages from your own corpus at query time and asks the model to answer using only those passages — which is the single most effective way to reduce invented facts.

How it works

Documents are chunked and embedded into a vector index. At query time the question is embedded, the nearest chunks are retrieved (often re-ranked), and they are placed in the prompt with an instruction to answer only from that context and cite it.

Use cases

Internal knowledge assistant

Answer staff questions from policies, runbooks and wikis, with links to the source.

Corporate

Customer-facing docs search

Let users ask questions of product documentation and get cited answers.

SaaS

Legal and compliance research

Search a contract or regulation set and quote the governing clause.

Legal

Support deflection

Resolve common tickets from the knowledge base before they reach an agent.

Support

Benefits

  • Answers reflect your current documents, not a training cut-off.
  • Citations make every answer auditable.
  • No retraining needed when the knowledge base changes.
  • Access control can be enforced at retrieval time.

Limitations

  • Answer quality is capped by retrieval quality — bad chunks, bad answer.
  • Chunking strategy materially changes results and needs tuning.
  • Multi-hop questions spanning several documents remain hard.
  • Keeping the index in sync with the source of truth is real ongoing work.

What to look for when choosing a tool

  • Citations linking back to the source document
  • Permission filtering applied during retrieval
  • Re-ranking, not just nearest-neighbour lookup
  • Incremental re-indexing when documents change
  • Evaluation tooling to measure groundedness over time

FAQ

Does RAG eliminate hallucination?
It reduces it substantially but does not eliminate it. A model can still misread a retrieved passage, so citations that a user can click through to the source remain essential.
RAG or fine-tuning?
RAG for knowledge that changes or must be cited. Fine-tuning for style, format and task behaviour. They solve different problems and are frequently used together.
How do I keep permissions correct?
Filter at retrieval time using the asking user's access rights. Never retrieve first and hope the model declines to quote a document the user cannot see.