Agent memory
Information an AI agent writes down in one session and reads back in a later one.
The memory tool lets Claude store and retrieve information across conversations in a directory of memory files.
Anthropic's page adds that an agent "records what it learns in memory files and reads them back on demand", which keeps the working context small. In its design the application, not the model, stores the files.
How it differs from its neighbours (the site's reading, not Anthropic's):
- Memory is context the agent keeps and is told again later. It is read, not run.
- A hook is code that runs at a fixed point whether or not the model wants it to.
- A tool call is the model asking for one action. Memory is often written and read through tool calls.
- Deep research is a task an agent does for you once. Memory is what any task leaves behind.
Mackenzie's way of saying it (2026-11-06): memory is "a special ritual around adding context to an agent".