ChatGPT forgets. Here is why that matters and what to do about it
ChatGPT's memory is limited and lives on its provider's servers. Here is what that costs you over time, and how an assistant with memory you own changes the maths.
ChatGPT is a genuinely clever conversationalist. Ask it anything and the answer is usually good. But spend a few weeks using it for real work and a pattern shows up: it does not really remember you. Its memory is limited by design, and what it does keep lives on its provider’s servers, not yours.
This is not a complaint about the model. It is a question about what kind of relationship you can have with a tool that starts most conversations close to scratch.
What “limited memory” means in practice
ChatGPT can hold some facts about you and recall a thread within a session. Outside those bounds, it forgets. The context window fills up and old detail drops off. Memory features are capped, summarised, and stored on the provider’s side. You do not decide what is kept, you cannot point to where it lives, and you cannot take the accumulated picture with you.
For a one-off question, none of this matters. For ongoing work, it adds up. You re-explain your clients. You re-paste the brief. You remind it of the decision you made last Tuesday. The assistant is brilliant each time and a stranger every time.
The cost is not the model. It is who owns the memory.
The instinct is to fix this with a better model or a longer context window. That misses the real gap. The difference between ChatGPT and an assistant who knows your work is not intelligence. It is ownership of memory.
Compare them on the axis that actually matters:
- Who owns the memory. With ChatGPT, the provider does. With an assistant whose memory lives on your server, you do.
- Memory across sessions. ChatGPT remembers you in a limited way. An owned assistant remembers your whole practice, out of the box.
- Take it when you leave. From ChatGPT, no. From your own server, yes, the whole thing.
- Where the data lives. With ChatGPT, in the provider’s regions. With intakto, on your server, in the EU.
Put plainly: ChatGPT is a very clever stranger, clever every time but never yours. The alternative is an assistant who knows you and knows you better every month.
What to do about it
You have three honest options.
Keep using ChatGPT and accept the re-explaining. For light, occasional use this is fine. There is no shame in using the right tool for small jobs.
Build your own assistant with memory. The memory is yours once you have built it, but that is weeks of setup: a server, Docker, access keys, a memory engine to maintain. One wrong step wipes the knowledge base.
Use a managed assistant whose memory you own. This is what intakto does. Hermes runs on a server that is yours, hosted in the EU, with lasting memory of your clients, projects, and decisions. You get the ownership of the self-built route without the weeks of setup. You also choose the AI model and can switch it later, because the memory lives with you and moves when you do.
The short version
ChatGPT forgetting is not a bug to wait out. It is a design choice about whose servers your context lives on. If your work benefits from an assistant who remembers, the question is simply where that memory should live: with a provider, or with you.
intakto is one plan, EUR 29 a month, with a 14-day money-back guarantee, and you can take the whole server with you whenever you want. See the full comparison with ChatGPT on memory if you want the detail.