CrewAI memory

CrewAI memory is the system that allows CrewAI agents and crews to retain and retrieve information across tasks and sessions — including short-term memory for recent context, long-term memory stored in an external database, entity memory for tracking specific entities, and contextual memory that combines these sources for relevant retrieval.

Memory types in CrewAI

CrewAI implements a multi-tier memory system. Short-term memory stores the most recent interactions and task results within a crew run, making recent context available to agents that need it without requiring them to receive every previous task output explicitly. Long-term memory persists information to an external storage backend that survives across crew runs — allowing an agent to recall relevant information from previous sessions when working on a related task days or weeks later. Entity memory tracks specific named entities — people, organizations, projects, concepts — maintaining accumulated facts about them across interactions. Contextual memory is a retrieval layer that queries all available memory sources and returns the most relevant information for a given task context.

Configuring and enabling memory

CrewAI memory is disabled by default and must be explicitly enabled when creating a crew. Short-term and entity memory are enabled with a flag; long-term memory requires configuring an external storage backend — typically a vector store or database that CrewAI can write to and query. Once enabled, memory collection is largely automatic: agents' outputs and the information they process are stored without explicit save calls in the agent code. Retrieval is also automatic: when an agent begins a task, the memory system retrieves relevant context from available memory sources and includes it in the agent's context. Custom memory configurations allow control over storage backends, retention policies, and retrieval parameters.

Privacy and data management considerations

Enabling long-term memory means the crew stores information about interactions in persistent external storage. For applications that process personal information, this creates data handling obligations: what is stored, for how long, who can access it, and how it can be deleted. CrewAI does not enforce data retention policies — those are the application developer's responsibility to implement through the storage backend's capabilities. Memory stored about individual users may be subject to privacy regulations including the right to erasure, which requires a path to identify and delete all stored data about a specific individual — a capability that must be designed into the storage architecture before memory is enabled for user-facing applications.

CrewAI memory — FAQ

Does CrewAI memory work across different crew runs on different days?

Short-term memory exists only for the duration of a single crew run. Long-term memory persists to external storage and is available across runs, including runs days or weeks later, as long as the storage backend retains the data. Entity memory also persists across runs when backed by external storage. For applications where cross-session recall is required — a personal assistant that remembers past preferences, a support agent that recalls previous issues — long-term memory with an appropriate storage backend is the required configuration.

How does CrewAI memory interact with LLM context windows?

CrewAI memory retrieves relevant information and includes it in the agent's context at task execution time, which means retrieved memory content counts against the model's context window limit. Long-term memory using vector retrieval selectively includes only the most relevant stored content rather than loading all stored memory, which manages context window usage better than verbatim buffering approaches. For agents with extensive memory and long tasks, monitoring context window utilization prevents situations where memory retrieval consumes most of the available context, leaving insufficient space for the agent's task content.