Documentation: Long-term Memory

Memo can store user preferences/profile info in a local memory file and inject it into future system prompts for more consistent behavior.

Memory File Location

Default:

  • ~/.memo/Agents.md

You can relocate it with MEMO_HOME (see Configuration).

What `save_memory` Should Store

Suitable:

  • Language preferences: for example, "User prefers Chinese responses"
  • Technical preferences: for example, "User prefers TypeScript" or "leans functional style"
  • Output preferences: for example, "Keep answers concise, conclusion first"

Not suitable (do not store):

  • Project structure/business logic/repo details (put these in project AGENTS.md)
  • Secrets/tokens/personal sensitive information

How It Is Used in Practice

Usually you only need to state your preference in chat; the model may call save_memory when appropriate.

If you want explicit memory capture, say something like:

  • "Please remember: I want all responses in Chinese and as concise as possible."