Skip to content
Agent Fleet

Choose which memory your agents use

There are three things an agent “remembers”.

  • Instruction files: rules you or your team write down (CLAUDE.md, AGENTS.md, ⚙ Settings > Agent instructions)
  • The agent’s own memory: what each kind of agent writes down by itself (Claude Code’s auto-memory, Codex’s memories, Muse Code’s memory)
  • Agent Fleet memory: the memory every kind of agent shares (⚙ Settings > Agent memory)

How it works

Section titled “How it works”
Instruction files The agent’s own memory Agent Fleet memory
Who writes it You (your team, for the repository’s files) The agent itself, while it works Agents, with af’s memory_* tools. You can write it in the memory view too
Where it lives The repository (committed) and each agent’s instruction file Each kind’s own files (Claude Code’s MEMORY.md and the rest, Codex’s ~/.codex/memories/) Inside the workspace, in a store Agent Fleet manages
Which agents read it The kinds that read that file. Agent instructions reach most kinds Only the kind that wrote it With the switch on, every kind that has af’s memory tools
Where you see or undo it in the Console ⚙ Settings > Agent instructions ⚙ Settings > Memory snapshots (Claude Code, Codex and Muse Code) ⚙ Settings > Agent memory > Memories and Changes
When the switch is on A short fixed note on the memory tools goes in beside your text Switched off for the kinds that support it (next table) New sessions read and write it

The switch is Let sessions use Agent Fleet memory under ⚙ Settings > Agent memory > Settings (off by default). How to turn it on is in Set how memory is used.

What the switch does to the agent’s own memory

Section titled “What the switch does to the agent’s own memory”
Kind Switched off while on? Imported when switched on? Written back when switched off? Kept in snapshots?
claude Yes (new sessions and the assistant chat) Yes, by itself Yes. It stays off until that has finished Yes
codex Yes (features.memories). Switching off puts back the earlier value No No Yes (once Codex’s memories folder exists)
muse Yes (sessions started from then on) Not by itself. Import from muse brings it in by hand No No
agy, copilot, cursor, opencode, kiro, lcpp Agent Fleet does not switch it off No No Agent Fleet keeps no history of it

Whether to turn Agent Fleet memory on

Section titled “Whether to turn Agent Fleet memory on”
Turn it on when Leaving it off is fine when
You use several kinds of agent You use only one kind of agent
You hand off to another kind, or start child sessions of another kind There is no need to share what was learned with other kinds
You want to read and correct what was learned in the Console The agent’s own memory is enough
You want to use it from your own Claude Code or claude.ai too (Use Agent Fleet memory from outside Agent Fleet is a separate switch)
Section titled “Related”