Agent Memory
Agent Memory
A mechanism that lets an AI agent remember past conversations and actions to inform what it does next.
In Simple Terms
Agent memory is the mechanism that lets an AI keep track of past exchanges and actions it has taken. For example, a travel-planning AI assistant might remember that you mentioned disliking spicy food and use that preference when making its next suggestion. It also comes into play when an AI picks up right where it left off, so you don't have to explain the same thing over and over again.
Behind the Name
"Agent Memory" combines "Agent," which in English refers to someone who acts on behalf of another, and "Memory," the ability to store and recall information. Together, the name points to the mechanism that lets an AI acting autonomously, much like a human, remember past events and conversations with users. It's a key feature for helping AI support people more intelligently.
Take a Closer Look!
Agent memory is the mechanism that lets an autonomously acting AI remember past conversations and task results, then put that information to use in future actions.
With this in place, an AI doesn't just follow a single one-off command — it can respond naturally based on everything that's happened so far.
Put simply, agent memory comes in two types: "short-term memory" and "long-term memory." Short-term memory temporarily holds onto the flow of the current conversation, similar to how a person stays focused on the task right in front of them.
Long-term memory, on the other hand, is the mechanism that stores past conversations and user preferences in a database so they can be recalled whenever needed.
This setup lets an AI respond flexibly to each user based on their past exchanges. For example, by loading the progress of a previous task from a database, it can smoothly carry out the next instruction.
Agent memory doesn't work by directly training or rewriting the AI model itself — instead, it keeps consistency across conversations and tasks by having the AI reference externally stored information whenever it's needed.