Glossary · AI
Retrieval-augmented generation (RAG)
Also called: RAG, grounding, retrieval
Retrieval-augmented generation (RAG) is a technique in which relevant information is looked up from your own data at request time and added to the prompt, so the model answers from those facts rather than from memory alone.
A model only knows what it was trained on and what you put in the prompt. RAG is the second part done systematically: for each request, retrieve the few pieces of your data that matter (the customer's last orders, the refund policy, the product spec) and include them, with an instruction to answer only from that material.
The usual pipeline
- Index your documents (often split into chunks and converted into embeddings for similarity search).
- Retrieve the chunks most relevant to the question.
- Generate the answer with those chunks in the prompt, ideally citing which ones it used.
You may not need the full thing
Vector databases and embeddings are one way to retrieve. For most business automations, retrieval is simpler: look up the record by ID, fetch the policy by category, or read the rows for this customer. A precise lookup beats a fuzzy similarity search whenever you know what you're looking for.
What RAG doesn't fix
It reduces invention; it doesn't stop it. Tell the model to say "not in the provided information" when the answer isn't there, keep retrieved passages short (they compete for the context window), and validate anything the answer claims before acting on it.
In Fluxo Flow
Fluxo Flow doesn't sell a vector database. Its Retriever node looks up relevant entries from Memory or static data and passes them to an AI step; HTTP Request and Integration steps fetch records from your own systems by ID. A founders example on our site: a webhook question → Retriever finds relevant entries → an Agent answers from those entries only → a Validator checks the answer against a JSON schema.
Read next
Related terms
- Context windowA context window is the maximum number of tokens a language model can take into account in a single request, counting the instructions, the input data and the reply it writes.
- LLMAn LLM, or large language model, is a neural network trained on large amounts of text that generates text by repeatedly predicting the next token, which lets it summarize, classify, extract and draft.
- Structured outputStructured output is a model response in a predefined, machine-readable format, typically a JSON object matching a schema, so the next steps of a workflow can use the fields directly instead of parsing prose.
- RPA (robotic process automation)Robotic process automation (RPA) uses software bots that operate applications through their user interface (clicking, typing and reading screens) the way a person would, typically to automate systems that have no usable API.
- RBAC (role-based access control)Role-based access control (RBAC) is a way of managing permissions in which access rights are attached to roles (such as Admin, Editor or Viewer) and people get rights by being assigned a role, rather than permission by permission.