Memories: what it remembers
Two things carry context between sessions, and only one of them is yours to control.
Memories let ChatGPT and Codex carry useful context from earlier work into later work. The web uses ChatGPT memory; local Codex clients use a separate local memory store with their own controls.
The rule from the documentation
Keep required team guidance in AGENTS.md or checked-in documentation, and treat memories as a helpful recall layer rather than the only source for rules that must always apply.
| AGENTS.md | Memories | |
|---|---|---|
| Who writes it | You | The tool, from your work |
| Where it lives | In the repository | In a local store, or your account |
| Reviewable | In a diff, like any file | Only if you go and look |
| Use for | Rules that must hold | Recall that helps |
The practical version: anything a colleague would need to know belongs in a file they can read. Anything that just saves you retyping can be a memory.
Controlling it
In the desktop app /memories decides whether a chat may use local memories and whether it may contribute to future ones. It is worth knowing where that switch is before a session picks something up you would rather it had not.
- Look at what your local memories currently contain.
- Move anything in there that a colleague would need into
AGENTS.md.
Slow is fine. Stopping is the only problem.