Menu

I was building an AI memory layer for PR work. Then MemoryBox arrived.

Alan Wallace is testing MemoryBox, a desktop AI memory application for carrying approved context, documents and working preferences across PR and communications workflows.

PressContact Blog · 3 minute read

I was building an AI memory layer for PR work. Then MemoryBox arrived.

I was building an AI memory layer for PR work. Then MemoryBox arrived.

By Alan Wallace

Every AI tool I use for communications work starts a new project knowing nothing about the last one.

I use ChatGPT, Claude and Gemini, depending on the job. Each one needs my editorial rules, the client background and my earlier research explained again. When I switch models, the explaining starts over.

So I started building around the problem. My research sits in Google Drive, my structured records in Notion and my project files on my own computer. I was writing tools to carry that material from one model to the next.

Then MemVerge invited me to test MemoryBox, its desktop memory application, now in beta.

I am in the middle of that test. I like what I see, and I am using it to improve the tools I had already started.

What it does

MemoryBox keeps documents, chat history and writing preferences on your own computer. It then supplies that background to whichever model you are using.

Its website lists a browser extension for ChatGPT, Claude and Gemini. It also lists connections to Google Drive, OneDrive, SharePoint, Dropbox and Box.

Three parts matter for communications work.

Documents index files and folders over a long client relationship, which covers old releases, messaging documents, crisis plans and coverage reports. All of it stays useful only if you can find it.

Personas hold working rules. Mine are short: AP style, plain sentences, no unsupported claims and no manufactured quotes. I have typed those rules into chat windows more times than I can count.

My Spaces keeps projects apart. One client's positioning should never surface in another client's draft.

There is also an experimental App Builder. That is where I am spending most of my time.

What I am doing with it

The tools I was already building are the test.

SignalRoom is one of them. It organizes company and competitor intelligence in cybersecurity, AI, and data center infrastructure.

A tool like that depends on history. A new announcement means little unless the system knows what the company said last year. That is the part I was building by hand.

Notion stays. It holds my structured company records, and MemoryBox does not replace it. I have not yet confirmed how cleanly the two connect.

What I do not know yet

This is a beta, and I have not finished testing. I have no retrieval scores to report and no finished app to show. I also have not verified what leaves my computer when an outside model handles a request. MemoryBox says workspace data stays on local hardware. For an agency holding client material, that needs checking before any confidential file goes in.

What I am measuring

  • Does it find the right passage in several years of documents?
  • Does a persona hold its style across two different models?
  • Does one client space stay out of another?
  • Does it cut the time I spend repeating instructions and correcting drafts?

I do not need another AI tool to draft a press release. I need tools that know what the company has already said, follows my editorial rules, and works the same way whether I am using ChatGPT, Claude, or Gemini.

I will publish the results when the testing is done.

Links and disclosure

Disclosure: MemoryBox offered me the opportunity to evaluate the product. This is a report about what I have seen so far. I am not finished, and there will be more to come.