Enterprise search is one search box that covers every application a company uses, and shows each employee only what that employee is allowed to see. It is also called workplace search. Most enterprise search products now include an AI assistant, which answers a question from the search results instead of only listing documents.
It works in four steps. First, connectors link the search to each application: email, chat, the wiki, the shared drive, the ticket system and the sales system. A connector is a small program that reads one application's data. Second, the search copies the text of every document into an index, together with each document's permissions, meaning the list of who may open it. An index is a large table that records which words appear in which documents, so a search can find them quickly. Third, when an employee searches, it finds the matching documents and removes every one that the employee may not open. It confirms who the employee is through single sign-on (SSO). Fourth, it ranks what remains, by how well each document matches and by signals such as how recent it is and whether the employee's own team wrote it.
Consider a new regional manager at a paint company in Nagpur. In his second week, he asks: "What did we decide about the dealer discount for the monsoon season?" The search finds the meeting notes on a page in the company wiki. It finds a chat thread in which the head of sales confirmed the decision: a 3% discount for dealers who order paint worth ₹5 lakh or more in a month. It also finds a spreadsheet of each dealer's margins. But that spreadsheet may be opened only by the finance team. So the manager does not see it, not even its title. The AI assistant then answers in two sentences, with links to the notes and the thread.
The AI answer works in two stages. The search first picks the most relevant passages. Then it gives them to a large language model (LLM), which writes the answer from those passages only. This is called retrieval-augmented generation (RAG).
The hardest engineering problem is permissions. A document's permissions change often. If the index does not update them quickly, the search may show someone a document they should not see. So enterprise search must follow the rule of least privilege exactly: each person sees only what they could already open in the original application.
Other products use the same search. An AI service desk agent uses it to find the answer to an employee's question in the company's policies. Support staff use it inside help desk software to find how a past problem was solved. Some products also build an enterprise knowledge graph, which records how tickets, code and documents are linked.
Enterprise search is different from web search. Web pages are public, and they link to each other a great deal, so links help to rank them. A company's documents are private, have different permissions for each person, and seldom link to each other.