Memory (in AI systems)

Information that persists between separate AI conversations or sessions, giving the model continuity it would not have on its own.

Every time you start a new conversation with an AI model, it begins with a blank slate. It has no recollection of what you discussed yesterday. Memory is the layer that fixes this: your application stores key information between sessions (conversation summaries, user preferences, factual records) and loads the relevant pieces into each new model call. Think of keeping a file on a colleague you meet with weekly. The colleague forgets every conversation between meetings, so you hand them your notes at the start and they pick up where you left off.

Builder example

A personal finance assistant that forgets the user's income and goals every session forces the user to re-explain their situation each time. Memory lets you store those details and reload them, making the experience feel continuous. The design challenge: deciding what to remember, how long to keep it, and how the user can review or delete it, especially when the stored information is sensitive.

You spend twenty minutes explaining your project to an AI. The next morning you open a new chat and it asks 'how can I help you today?' with zero recollection.

Save key project context, preferences, and decisions to a memory system that reloads in new sessions. Let users see and edit what is stored.

Common confusion: Memory and context window are different things. The context window is everything loaded into one call (the papers on your desk right now). Memory is the filing cabinet: it persists between calls, and your application chooses which files to pull out for each conversation.