Sopheons
Every conversation with an LLM begins with a small act of repetition.
You explain who you are, what you are working on, how you like things done and what has already been decided. The longer and more specific the work becomes, the more frustrating this is. The AI may be capable, but it does not really know your history.
This is the problem I am trying to solve with Sopheons.
Sopheons keeps useful knowledge from your files, projects and previous conversations in a memory that you control. When you begin a new task, it can prepare a focused brief containing only the relevant context. You can then take that brief to ChatGPT, Codex, Claude or another LLM.
The important point is independence. Your memory should not belong to one AI provider, and it should not be trapped inside the application that organised it. Models will change. Services will improve, become more expensive or disappear. Your accumulated knowledge should survive all of that. With Sopheons, the files remain yours and the LLM remains replaceable.
Why should an ordinary user bother?
Because the benefit is not limited to remembering preferences such as tone, language or formatting. Over time, the memory can also preserve *how you approach particular kinds of work.
Imagine starting a new application. Sopheons can help turn what you already know—its purpose, structure, decisions, preferences and next steps—into a detailed project blueprint. You can use that blueprint yourself, share it with other people, or turn it into a much richer prompt for an LLM.
There is nothing magical about this. You choose the files and conversations you want to share. Sopheons extracts the useful memories, lets you review them and helps place the relevant ones into the prompt. That prompt might say: publish my code to GitHub when it is ready; propose updates to my memory after this task; keep emails to John formal but write to Andy in a more personal style; remember that John prefers afternoon meetings.
The LLM still has to think, and memory is not a substitute for judgement. It simply starts with a clearer blueprint, better instructions and more of the experience you have already accumulated.
The same idea applies beyond coding. A writing brief can carry your preferred voice and the corrections you have made before. A research brief can preserve trusted sources, open questions and rejected assumptions. A project brief can record what has been completed, what remains uncertain and what should happen next.
That is what I find most valuable: not merely a machine that remembers facts about me, but a portable body of experience that helps different machines work with me more intelligently.
Sopheons is my attempt to make that memory useful, reviewable and—most importantly—mine.
Originally published on Substack on 20 August 2026.