Every product runs offline. One-time payment — no subscription, no account, no telemetry.
All apps Documents & LegalBusiness & FieldEngineering & DesignMedia & AudioSystem & Privacy BundlesGuides DeliveryWhy local Windows appsAndroid apps
Guide

How to run e-discovery on an air-gapped network

A practical walkthrough for collecting, indexing, reviewing and producing case documents on a machine with no network connection — including what breaks and how to work around it.

Updated 2026-11-18·Covers local-first deployment

Most e-discovery advice assumes the data can go to a server somewhere. For a growing share of matters it cannot: defence work under an ITAR clause, investigations inside a classified environment, litigation where a client's contract names the countries its data may not leave. The work still has to happen. Here is how it happens without a network.

Start with the constraint, not the tool

Write down what is actually prohibited before choosing anything. There is a large difference between "no internet connection" and "no removable media leaves the room". The first is a network configuration; the second determines how you get the collection in and the production out. Getting this wrong is expensive — discovering at production time that the only export path is a USB port someone disabled is a bad day.

Bring the collection in

Collection is the step that most often breaks an air-gapped plan, because the tools that collect from custodian machines are themselves network tools. Three patterns work:

  • Collect on the edge, transfer on media. Run the collection on a machine that can reach the sources, hash the output, write it to approved media, and carry the media across the gap. Keep the hash manifest with the media and verify it on the far side before you open anything.
  • Ask for native files. For a small matter, the custodian's own export — PSTs, folders, a zip of the document store — is often sufficient and avoids a collection tool entirely.
  • Image the source. Where the file system itself is the evidence, take a forensic image and work from a copy. This is slower and larger but it removes arguments later.

Index on the machine that will review

On an air-gapped workspace, processing and review collapse into one machine. That changes the sizing question: instead of "how many documents can the server hold", it becomes "how many can this workstation index and keep searchable". In practice a modern laptop handles a workspace in the tens of thousands of documents comfortably; six figures is where you should be looking at a server platform and accepting that it will be a separate, connected system with its own controls.

Split by custodian or by date range rather than forcing everything into one workspace. Two workspaces that each open instantly beat one that takes twenty minutes to load.

Review without a collaboration server

This is where teams feel the loss most. On a connected platform, a dozen reviewers code the same set with shared tags and quality control. Offline, review is either single-reviewer or it is distributed and reconciled.

The pattern that works: one reviewer per workspace, with a coding schema agreed in advance and written down. Export coding decisions as a table at the end of each session and merge them on a control machine. It is slower than live collaboration and it requires discipline about the schema, but for a matter with three reviewers and 8,000 documents it is entirely practical.

Produce with an integrity record

A production handed to opposing counsel needs to answer one question: is this the same document they collected? A hash manifest over the collection, carried through processing, and a hash chain over every action taken gives you that answer without asking anyone for a log extract. Bates numbering, page stamping and a load file make the production usable; the hash chain makes it defensible.

What genuinely does not work

Be honest with yourself about the edges:

  • Machine translation and cloud AI review. Both require the documents to leave. There is no air-gapped equivalent that performs as well.
  • Live collaborative review. Distributed and reconciled is not the same thing.
  • Very large matters. Above the low tens of thousands of documents, a workstation is the wrong shape of computer and no amount of software design fixes that.

The point of an air-gapped e-discovery workflow is not that it matches a server platform feature for feature. It is that the matter can proceed at all when the alternative is not proceeding.

Questions

Can I use AI review on an air-gapped machine?
Only with a locally hosted model, which means a GPU workstation and a model small enough to run there. Quality is materially below cloud review. Treat it as triage, not as a substitute for review.
How do I get the production out?
On approved media, with a hash manifest, through whatever transfer procedure the environment requires. Confirm the destination can read the format before you produce — a load file nobody can open is not a production.