--- title: "Enterprise Search for AI (RAG) – Lookeen Server" url: https://lookeen.com/solutions/rag lang: en description: "Use Lookeen Server as a secure RAG layer: make corporate knowledge from email, files and Microsoft 365 available to your own AI agents through an API or MCP." published: 2026-10-23T10:00:00+00:00 modified: 2026-11-12T10:00:00+00:00 --- # Enterprise Search for AI (RAG) Give your AI applications access to corporate knowledge. Search email, files and Microsoft 365, retrieve relevant passages and read the source documents through one API. [Discuss your integration](https://lookeen.com/contact) [View the RAG API](https://lookeen.com/solutions/rag/api) On-premises·API & MCP·Permission-aware // Retrieve relevant context POST /v1/rag { "query": "Project Phoenix risks", "user": "alex@example.com", "limit": 5 } ✓ Relevant documents, ready to read ✓ Stable document IDs for citations ## Turn enterprise search into your retrieval layer RAG means retrieval-augmented generation: before a language model answers, it receives relevant content from reliable sources. Lookeen handles indexing, retrieval and permission filtering; the model, prompt and agent logic remain your choice. ### Connect existing data File servers, network drives, Exchange, Microsoft 365, SharePoint, OneDrive and PST archives feed one central index that stays up to date. ### Supply focused context Keyword search returns matching documents with a preview. Semantic search adds relevant passages. Read a document's text when your application needs more context. ### Keep sources traceable Every result includes a document ID for reading and citations, with its name, source, type and modification time where available. IDs remain stable across reindexing and content updates, but change when a document moves because they encode its path. Semantic passages include page numbers when known. ## Semantic search: find meaning, not just matching words Ask a question in natural language and find passages that relate to its meaning—even when the source uses different wording. For example, a search for “Why was the customer migration postponed?” can find a passage about a delay pending security approval. This helps an AI application find useful context when the question and the original document do not use the same terms. Lookeen uses an embedding model to represent document passages and the query as vectors, then ranks passages by similarity. Results include source information and page numbers when available, so your application can trace a passage back to its document and read more context. Similarity is not a guarantee that a passage answers the question; your application should evaluate the results before using them. Keyword search remains the default and is useful when exact terms matter. To search by meaning, enable semantic search and set `mode: "semantic"` in the [RAG API](https://lookeen.com/solutions/rag/api). You can configure a local embedding endpoint or use an external provider, depending on your data-handling requirements. ## Two ways to connect applications and agents Your own application or an off-the-shelf AI assistant: Lookeen delivers the right sources through an open interface. ### HTTP API Send a `query` to `POST /v1/rag`. Keyword search is the default; set `mode: "semantic"` to search by meaning. Add an optional `user` as a user principal name (UPN) or email address to apply that user's permissions. Filter by name, source, item type, extension, path or date, then read a result with `GET /v1/rag/documents/{id}`. [Complete API reference](https://lookeen.com/solutions/rag/api) ### Model Context Protocol (MCP) Through MCP, Lookeen can be exposed as a tool to compatible assistants and agents. An agent searches corporate knowledge when needed, receives sources and incorporates the findings into its own workflow. [Ask about MCP integration](https://lookeen.com/contact) ## Retrieval without a prescribed answer The RAG API never generates an answer. Keyword search makes no AI calls; semantic search uses the configured embedding model. Choose your own answer-generation provider, system prompt and citation format. For complete conversational research, [Lookeen AI](https://lookeen.com/ki) is available alongside it. ## Built for corporate data Handing company knowledge to AI systems takes control over where it lives, how much is shared and who gets access. ### Inside your infrastructure Lookeen Server and its index run on your network. For isolated semantic search, configure a local embedding endpoint. External embedding providers receive extracted text chunks during indexing and query text during retrieval; review their data handling before enabling them. ### Control result count `limit` controls how many documents are returned in keyword mode or passages in semantic mode: 1–50, default 10. The read endpoint provides the full extracted text, separated by page. These are not byte or token limits; your application controls how much text reaches the language model. ### Optional user filtering Pass a UPN or email address in `user` to search with that user's document permissions. Lookeen looks up the user and resolves groups and delegations. Omit `user` to search without a user-specific permission filter. Use the same permission context when reading a result's source text. ## Typical uses Where Lookeen as a retrieval layer makes the difference today. - **Internal assistants:** answer questions about projects, customers, policies and earlier communications. - **Agentic workflows:** search for evidence before acting and pass cited sources into the next step. - **Business applications:** provide relevant email and documents inside CRM, service, legal or compliance processes. - **Your own AI platform:** add local enterprise retrieval to existing models and orchestration systems. ## Frequently asked questions What the RAG API returns, how it searches and what your application can do with it. Does the RAG API generate an AI answer? No. It returns matching documents with excerpts and, in semantic mode, relevant passages. Your language model or agent performs generation. Keeping the steps separate makes control, evaluation and model changes easier. Can I search by keyword or by meaning? Yes. Keyword search is the default. Set `mode: "semantic"` to search by meaning and receive relevant passages. If semantic search is not enabled, the API returns an error. Can my application read a matching document? Yes. Use the document's `id` with `GET /v1/rag/documents/{id}` to retrieve its extracted text, grouped by page. The [API reference](https://lookeen.com/solutions/rag/api) includes search and read examples. Related: [Lookeen AI](https://lookeen.com/ki) · [Lookeen Server](https://lookeen.com/solutions/enterprise-search/lookeen-server) · [Local AI](https://lookeen.com/use-cases/local-ai) · [Data protection & GDPR](https://lookeen.com/use-cases/data-protection-gdpr) ## Put your corporate knowledge to work for AI Talk to us about data sources, permissions, API or MCP integration and deployment in your infrastructure. [Talk to us](https://lookeen.com/contact) [API reference](https://lookeen.com/solutions/rag/api)