Memory for agents: so they don't start from scratch

What you'll see in this lesson

1. Why your agent forgets everything

   An agent only knows what's in its context: if you clear the session, it starts from scratch unless it saved something outside.

2. Engram: local memory for coding agents

   Engram, by Gentleman Programming, is a single program that gives persistent memory to almost any agent through MCP.

3. claude-mem: automatic layered memory

   claude-mem records what the agent does without being asked and feeds it back little by little in the next session.

4. mem0: memory for your own apps

   If you build your own agent or chatbot, mem0 adds per-user memory with two functions: add and search.

5. Local or cloud memory

   Locally, your data never leaves home; in the cloud, you share it across teams and people without setting anything up.

6. Beads: tasks that agents understand

   Beads, or bd, is a task manager for agents: it stores each task with its dependencies and tells them what they can do now.

7. Backlog.md: the board in markdown files

   Backlog.md stores each task as a .md file in your project and gives you a board to review them with your agents.

8. Several agents, one memory

   When several agents work on the same project, they must all read and write in the same places, without stepping on each other.

9. Summary: agents that don't start from zero

   Memory outside the context, retrieved on demand, and shared tasks that every agent understands.