Most people think their AI has a memory. It has two, and neither one is about you.
There are six kinds of memory worth knowing. Four come from cognitive science. Two are AI's own.
➡️Parametric memory is the knowledge baked into the model during training. It is why it can write and reason. It knows the world, not you.
➡️ Working memory is the current conversation, sitting in the context window. Fast, and gone the moment you close the tab.
That is what a plain chatbot runs on. Parametric plus working. Everything past that, it does not keep.
The four that actually make it feel like it knows you:
➡️Episodic memory: these are specific events like "You asked about the pricing page last Tuesday."
➡️Semantic memory: these are durable facts and preferences like "This user prefers Python over JavaScript."
➡️Procedural memory: how you like things done. "Check the budget first, then book the best price-to-performance option."
➡️External memory: your docs, codebases, notes, and tools, retrieved when relevant instead of pasted into every prompt.
What's interesting here: All four need storage that lives outside the model. The model cannot hold them on its own, and each platform that tries keeps its version locked inside its own walls.
So you rebuild your context in every new tab, in every new tool, forever.
The fix is not a smarter model with bigger context windows. It is a memory layer that sits outside all of them, holds the episodic, semantic, procedural, and retrieval pieces in one place, and carries them across ChatGPT, Claude, Gemini, and the rest.
That is the whole idea behind AI Context Flow. You teach it once. It remembers everywhere.