A private AI that answers from your own documents.

Yes. Engineers call it retrieval-augmented generation, or RAG. Instead of a general-purpose model answering from memory, the system retrieves the relevant passages out of your own files first, then writes the answer and cites the document it came from. Almost nobody searches for the acronym, so we describe it in plain language everywhere else on this page.

It is part of our custom software service. A private assistant on your documents is one of the things we build inside a larger system, alongside internal tools and integrations, and it is usually scoped in the same engagement. It has its own page because people ask for it by name, not because it is sold separately.

Only if the system calls their model, and you decide that before we build. The application, database and document index can run on your own server or cloud account; the case-management system we built for Healing Hands Ukraine, for example, keeps its database on the foundation's own cloud server. When the system does call an outside model, the text it sends goes to that provider, and the terms differ by plan. OpenAI does not train on API data by default and keeps abuse-monitoring logs for up to 30 days. Google uses content sent on the free Gemini API quota to improve its products and does not use paid-tier prompts that way. Anthropic does not train on its commercial products' inputs by default (provider terms as of September 2026). If nothing may leave your network, an open-weight model such as Llama or Qwen can run on a GPU server you own or rent. Before you sign off, we put in writing which model is called and what data it receives.

General AI tools like ChatGPT answer from their training data, which doesn't include your internal knowledge and is prone to hallucination. A private assistant on your own documents is constrained strictly to your approved material: it can only answer from your files, always cites its source, runs on your infrastructure, and if it ever calls an outside AI model, we tell you which one and what data it sees.

We can index internal documents and PDFs, wikis and knowledge bases (Notion, Confluence, SharePoint), CRM records, email archives, databases, and any other structured or unstructured data source your team uses. Everything is ingested into a private index on your infrastructure.

It can be, and you judge that on evidence rather than a label, because compliance is measured against your own obligations. In writing you get where the data sits, who can log in and with which role, how every change is logged, and each outside model with the data it receives. The Healing Hands system shows the access side: permissions are split by role and regional center, the clinical module lives in separate tables, and the audit log recorded 6,116 staff actions in the first 66 days in production. Patient records and client files carry their own rules; see how we handle them for medical practices and law firms.