Memory

An agent that forgets the last finding will re-research the same topic. Memory is how crews improve across a session and across days.

Scratch paper (short-term), a notebook you keep (long-term), and a CRM card per customer (entity). RAG is the library you walk to — not the notebook.

Visual Workflows

Start here — scroll inside each diagram frame to explore, then use + / − to zoom up to 200% if needed.

100%

Loading diagram...

Scroll inside the frame to explore · use + / − to zoom up to 200%

Key Takeaways

1.Memory is what the crew retains. Knowledge is what you retrieve from documents. They are not the same. Short-term memory is this run. Long-term memory persists across kickoffs. Entity memory is facts about people and things.
2.Flows can remember and recall across runs. Turning memory on is not a memory design.
3.Crew memory covers this execution and, if enabled, stored facts for later kickoffs. Scope memories so one topic or user does not leak into another.
4.Memory is not stuffing the window — you still budget context.

Learn elsewhere

→Agent Memory — Phase 5
→Context Engineering — Phase 6
→Knowledge and RAG

Real Example

Scenario

A research crew remembers that last week's brief already covered MCP sampling, so the next run starts from the gap.

What you would do

In CrewAI, apply Memory to this scenario: A research crew remembers that last week's brief already covered MCP sampling, so the next run starts from the gap. Identify the inputs, run the technique, validate the output, and note one thing you would monitor in production.

Commands

Commands to Remember

Memory ≠ RAG
Short / long / entity
Scope by user or run
Still budget context

Cheat Sheet

Quick recap

quick ref

•Notebook vs library
•Scope memories
•Not embed-all-PDFs
•TTL stale facts

Common Mistakes

✕Skipping evaluation for Memory before production
✕No logging or tracing around crewai memory steps
✕Ignoring cost and latency implications