Google adds private document search with Gemini AI for locked-down networks

Google has announced a preview of a private document search tool powered by its Gemini AI. It runs entirely inside an organization's own secure network, with no link to the public internet or to Google Cloud's global systems. It is aimed at national security agencies and other heavily regulated groups. Until now, those teams had to spend months building something like it by hand.

What it does

Staff upload files into a conversation and ask Gemini to find information, summarize, compare or translate. Every answer cites the files it came from. Conversations are saved, so a user can reopen an earlier chat and pick up where they left off. Google calls the approach retrieval-augmented generation: the AI looks up the relevant parts of your files first, then writes its answer from them.

What files it handles

• PDF, Word, PowerPoint and Excel files, plus TXT and HTML
• PNG and JPEG images, WAV audio and MP4 video
• Each file can be up to 100 MB, and each chat up to 500 MB

Gemini can read text, images, audio and video directly. That means the system does not need separate tools to turn speech into text or to read scanned pages. For spreadsheets, Gemini writes small pieces of Python code to calculate answers, such as an average salary by department. That code runs in a locked-down sandbox with no network access.

How privacy is handled

Users sign in through their organization's own identity system, and only approved groups can open the app. Each person's chats and uploaded files are private to that person. Google says prompts, answers, documents and the data the search uses to match files never leave the air-gapped network, and every model runs locally.

Data is encrypted when stored and when moving across the network. Google says the underlying platform has U.S. Department of Defense IL6 accreditation and is designed to support FedRAMP High security controls, a strict U.S. government standard.

Why it matters beyond government

Most readers will never use an air-gapped network. Still, the launch shows that private search over your own documents, with sources attached to every answer, is turning into a standard product instead of a custom project. Businesses that worry about where their data goes may start to expect the same from other vendors.

What is not known yet

The announcement does not give pricing or a date for the preview to become generally available. It also does not say which customers have tried it. Interested organizations are told to read the GDC air-gapped documentation and contact their Google Distributed Cloud representative. The tool runs on Gemini 3.5 Flash, using NVIDIA Blackwell Ultra hardware.