MILabs
Tools

PrivateGPT: private AI for questions about your own documents

PrivateGPT lets a business ask questions of its own documents on infrastructure it controls. Here is how MILabs sets it up and where it fits.

Updated 2026-01-20

What PrivateGPT is

PrivateGPT is an open-source project from Zylon for asking questions of your own documents. Its design keeps those documents on infrastructure the business controls. The source code and documentation are published in a public repository, so the points on this page come from there. The project runs the question-answering pipeline locally or in a private environment. Documents are split into sections and indexed for search, and a language model writes answers from the passages that match. Because the components can run on a company's own hardware or in a private cloud, the business decides where its files and questions are processed. Teams usually meet the tool when they hold a large body of written material they consult often, such as contracts, manuals, policies, past reports or client files. The answers depend on the quality of those documents, so setting one up is as much about choosing and maintaining the material as installing the software. Most of the work is in the preparation, and a clear list of collections is the first thing to settle. The sections below describe what the tool does, where businesses use it and how MILabs sets it up.

Zylon: official site

PrivateGPT in plain language

Question answering over documents works in two stages. First the files are prepared: text is pulled out, divided into sections of manageable length and stored so a search can find passages by meaning rather than exact wording. When someone asks something, the most relevant passages are retrieved and handed to a language model, which writes a reply from that material. The reply is only as dependable as the passages behind it, so the source documents matter more than the screen people type into.

Running this on a company's own infrastructure changes who can see the material. A cloud assistant typically sends each question and each retrieved passage to a provider's servers under that provider's terms. A self-hosted setup keeps those steps inside a machine or account the business administers, which can make conversations with clients, insurers and regulators about data location simpler.

The trade-off is responsibility. The business takes on work a vendor would otherwise carry, including applying updates, sizing hardware, managing access and choosing the language model. Self-hosting does not remove the need for judgement about answers, and it does not make a weak document set any stronger.

What businesses use it for

  • Answering staff questions from internal policies, procedures and onboarding material.
  • Finding specific clauses and dates across long contracts, leases and agreements.
  • Summarising technical manuals so field teams reach the right step quickly.
  • Querying past reports, tender files and project records on a network the business chooses.
  • Checking client files for one specific detail before a meeting begins.

Who it suits

  • Firms with large document libraries that need to stay on their own systems.
  • Teams that answer the same questions from policies or manuals each week.
  • Owners who have a named person to look after the setup over time.

How MILabs sets it up

MILabs starts with the owner, or the team lead who will use the tool day to day, and agrees which document collections the assistant should read and which stay out of scope. We then decide with the owner where the software will run, whether on a company machine, a private server or a private cloud account, so the deployment matches the business's own data rules. Next we connect the document folders or stores the business already keeps, and test the results against questions the team actually asks. Access is set so each person sees only the collections they are entitled to. We also write down who maintains the model, updates the software and reviews new documents. We work remotely from Sydney and run sessions by video call.

Builds it fits

Industries it fits

PrivateGPT in practice

Day to day, the arrangement works like a shared reference desk. A legal team might keep a folder of precedent agreements, schedules and standard terms, and an associate asks the assistant for every clause that sets a notice period. The answer points to the relevant passage, so the associate knows exactly where to look in the original document.

Access is the part that changes most often, and it follows the rules your team already knows. A finance team's collection may include payroll records and client tax files, which some staff see and others do not. The setup mirrors the folder permissions the business already has, and the owner updates them whenever someone changes role or leaves.

Upkeep is a small, steady habit. Each time a policy is revised, the new version goes into the searchable set and the old one comes out, so the assistant answers from the current wording. Many teams name one person to look after the collection, trying a sample of questions against the latest documents now and then. That habit, more than the software itself, keeps the tool sharp long after it goes live.

good questions, straight answers

Frequently asked questions

When is PrivateGPT the wrong choice?
PrivateGPT suits less well a business that wants a finished product with vendor support on a fixed schedule, or one with a small document set that a hosted assistant with simple file upload can handle. A team that expects every answer to be right without review of the library will also find it a weaker match. We compare material volume, data rules and the ongoing care the team can give.
Can my team use PrivateGPT without training?
Yes, for basic questions. Most people can ask through a simple interface without formal training, but reliable answers depend on knowing what the documents contain and how to phrase a question. We suggest a short walkthrough for the people who will use it, along with a written note on which collections are available and what the assistant can and cannot answer. Staff also check each answer against the original document.
Can MILabs build on top of PrivateGPT?
Yes. MILabs can install PrivateGPT, connect it to the document stores a business already uses, and build a layer around it, such as a staff question page or a workflow that routes answers to the right person. The owner agrees the build in advance, and it is documented so the business's team or an outside provider can maintain it. We are based in Sydney, work remotely and run sessions by video call.
What does a PrivateGPT setup need from me?
A setup needs a clear owner, access to the document collections you want searched, and a decision about where the software will run. We also need a sample of the questions your team asks, so we can test answers against real material. You will need to say which files are current, which are superseded and who approves changes to the library. Naming a person to look after it once running helps.
What does PrivateGPT not do?
PrivateGPT does not decide whether an answer is right or complete. It reads the documents it is given and writes responses from them, so a missing, outdated or poorly written file produces weaker answers. Its answers are only as sound as the material behind them. It is also no substitute for legal, medical or financial judgement. Each capability it offers depends on how the project is configured.
let's make something happen

Workflow automation: how we can help

Choose how you want to work on this with MILabs. Both options start with an enquiry.

Discovery is free, no obligation from the first chat.

Let us build it for you

We build it and look after it.

Build it yourself, we'll help

You build it with our direction, teaching and a weekly check-in.