For two years, we’ve been told that “Context Windows” are the answer.
Yesterday, Adam Danyal published a piece on AI memory lock-in. He argued that teams are bleeding time re-briefing tools because memory is trapped inside the vendor. He called for a Postgres-backed memory layer owned by the user.
He is absolutely right.
Only he calls it the future. For me? It’s 2023.
That Is When I Built xSyched
I built it because I was tired of shouting at a screen: “WE LITERALLY JUST TALKED ABOUT THIS.”
The “experts” said LLMs would eventually remember everything. They were wrong.
Even almost 3 years later, GPT, Claude, and Gemini only give you fragments. They still don’t really know who you are, or what your team already did.
A Different Premise
xSyched was built on a different premise: separating the “brain” from the “memory” while maintaining full control of that memory.
I happened to have used the same Postgres database that Adam proposes, to ensure memory outlasts the assistant:
- Context (Instructions, Project, Objectives, Items, Todos) defined ONCE.
- Persistent across years, not sessions.
- Shared across teams and groups.
- Zero “Memory Tax.”
Proof Left Behind
xSyched didn’t need any updates and I moved on to develop an AI Customer Support system for Marcelo Lombardo at Omie (xSyched User). But the code I left behind more than 2 years ago proves a hard truth the industry is only now realizing:
If you rely on the platform’s “memory,” you still don’t benefit from real memory, and it isn’t even yours. You’re just renting it.
The market is finally naming the problem I solved three years ago.
Building before the market is ready feels almost like explaining water to fish.
Stop Re-Explaining Yourself to Your Tools
How many hours did your team waste this week “reminding” an AI of things it should already know?
If the answer is more than zero, you aren’t using AI. You’re babysitting it.
xSyched.com is still available and totally FREE!
