AMAgentmemory
Home

Agent memory survey

Agent Memory Survey: What Builders Should Compare

Most agent memory surveys describe a broad research landscape. Builders need a narrower question: which memory pattern will actually help a team finish more work without creating privacy or correctness risk?

Memory types worth comparing

The strongest products combine several memory types. Episodic memory preserves what happened. Semantic memory preserves what is true. Procedural memory preserves how work gets done.

  • Episodic memory for session events and incidents.
  • Semantic memory for durable project facts.
  • Procedural memory for repeatable workflows.
  • Preference memory for operator-specific working style.

Operational questions

A survey is incomplete if it ignores governance. Teams need to know who can write memory, who can see it, how long it lives, and what happens when it is wrong.

  • Can users inspect and edit memory?
  • Can teams separate personal and shared scopes?
  • Can stale memories expire automatically?
  • Can every memory show its source?

Practical playbook

  1. List the decisions agents repeat every week.
  2. Map each decision to a memory type.
  3. Choose retrieval that fits the work: exact, semantic, graph, or hybrid.
  4. Run a two-week comparison against a no-memory baseline.

Related guides

Agentmemory githubClaude Code agent memoryAgent memory frameworkMem0