Agent memory is the stored state an AI agent carries between steps or sessions so it can continue work without being told everything again.
An agent is a model that runs in a loop: read the situation, pick a tool, act, repeat. Without memory, each loop iteration and each new session starts blind. Agent memory fixes that by recording what the agent has learned, decided and left unfinished. Short-term memory is the current context window. Long-term memory is anything written outside the window and read back later: a file, a database row, a vector store entry.
Good agent memory answers three questions on resume. What was decided? What is still open? What should happen next? Systems that only store raw chat logs answer none of them well, because the agent has to re-read everything. Systems that store structured notes (decisions, facts, open questions, todos) let the agent pick up in seconds. Frameworks such as Letta build memory into the agent runtime. Protocol-based approaches expose memory as tools any agent can call.
How Central Brain handles agent memory
memory_recall, memory_checkpoint and memory_compact, plus central_brain_ping to confirm the connection. A new session can call memory_recall to pick up where the last one stopped.