Retrieval-augmented generation (RAG) is a technique that answers a question by first retrieving relevant passages from a defined document collection, then having a language model write the answer from those passages.
A plain language model answers only from what it absorbed during training, which is fixed, generic, and impossible to inspect. A RAG system instead searches a specific library — a company's price book, vendor list, and procedures — pulls the passages that match the question, and passes them to the model as the material to answer from. Because the answer is built from named source passages, it can be traced back to the document it came from, and updating the library changes the answers without retraining anything.
Private ChatGPT alternatives for small business
Private or self-hosted AI is an arrangement where the AI model and the data it processes run on infrastructure controlled by the organization or its vendor, rather than being sent to a public AI service.
With a public AI service, every question and every document is transmitted to a third party that operates the model. Private or self-hosted AI keeps that processing inside a controlled environment, so business documents and the questions asked about them are not exposed to an outside model provider or used to train its models. For a small business this usually means a managed service runs the model on its behalf, not that the business operates servers itself.
What public AI services do with your business data
Multi-tenant isolation is the practice of keeping each customer's data and activity strictly separated inside a system that serves many customers from shared infrastructure.
A multi-tenant application stores many businesses' records in the same database and serves them from the same code. Isolation is what guarantees one business can never see, search, or retrieve another's content. It is normally enforced at more than one layer — for example a per-request identity check in the application plus row-level rules in the database — so a single missed check does not expose data across customers.
Is it safe to put your business data into ChatGPT?
AI data residency is where the data processed by an AI system is stored and where the computation on it takes place, stated as a geographic or organizational boundary.
Some contracts, regulations, and customers require that business data stay within a specific country or within systems a named party controls. With AI tools this is harder to pin down than with ordinary software, because a single request can pass through model hosting, logging, and safety-review systems in several locations. Knowing the data residency of an AI system means knowing every place a document or question travels once it is submitted.
Where your data goes when you use a public AI service
Grounding and citations
Grounding is requiring an AI answer to be derived from specific supplied source material, and citations are the references to that material shown alongside the answer.
An ungrounded assistant can state something fluently whether or not it is true, and gives the reader no way to check. A grounded assistant is constrained to answer from retrieved passages and marks each answer with the document it came from, so a reader can open the source and confirm it. Grounding also gives the assistant a defined way to say it does not know when the library holds no relevant passage, instead of guessing.
How to document the way you price jobs