how do codex memories work in the chatgpt desktop app?
codex memories are local files that let codex carry context from earlier chats into later ones. they are off by default. you turn them on in the chatgpt desktop app under settings > personalization, or in config.toml. openai's memories doc says chatgpt web uses chatgpt memory, and local codex clients keep a separate local memory store with their own controls. this is how it works as of november 2026.
download the ReMynd desktop app to search what you saw in a focused window, by app and date.
which memory does each app use?
openai documents local codex memories for the chatgpt desktop app, codex cli and the ide extension, and chatgpt memory for chatgpt web. here is what it documents for each.
| where you chat | memory controls openai documents | shares ~/.codex/config.toml? |
|---|---|---|
| chatgpt web | chatgpt memory, managed in settings > personalization > memory | no |
| chatgpt desktop app | local codex memories: enable codex memories per machine, /memories per chat | yes |
| codex cli | local codex memories: /memories per chat | yes |
| ide extension | the connected codex host's local memory store | yes |
for the mcp column, openai's mcp doc says the chatgpt desktop app, codex cli and ide extension share this configuration, and that chatgpt web doesn't read local codex configuration files.
how do you turn on codex memories?
you turn them on in the chatgpt desktop app or in config.toml, per openai:
- in the chatgpt desktop app: open settings > personalization and turn on enable codex memories. it applies to the selected machine.
- in config.toml: add the feature flag under
[features]:
``toml [features] memories = true ``
config basics puts your personal codex config at ~/.codex/config.toml, with optional project overrides in a trusted repo's .codex/config.toml.
how do codex memories work?
in the background, after a chat goes idle. openai says codex turns useful context from eligible prior chats into local memory files, skips active or short lived sessions, and redacts secrets from generated memory fields. memories may not update right away when a chat ends.
the configuration reference lists these defaults for the codex memories config:
| key | default |
|---|---|
memories.min_rollout_idle_hours | 6 (idle hours before a chat is considered) |
memories.max_rollout_age_days | 30 (maximum age, in days, of chats considered) |
memories.min_rate_limit_remaining_percent | 25% (codex rate limit left before generation starts) |
memories.generate_memories | true |
memories.use_memories | true |
memories.disable_on_external_context | false |
with disable_on_external_context set to true, chats that used mcp tool calls, web search or tool search are kept out of memory generation. in the desktop app, allow memories from tool-assisted chats chooses whether chats that use mcp tools or web search can contribute to future memories.
where are codex memories stored?
under your codex home directory, ~/.codex by default. openai says that when you enable memories in a fresh codex home, the cli uses $CODEX_HOME/memories_v2/ by default, and if $CODEX_HOME/memories/ already exists it keeps using that store. with no explicit memories.version, setting memories.dual_write = false keeps even a fresh profile on the older store.
in the desktop app, openai says to check both folders. openai calls these files generated state and advises against editing them by hand as your main control.
how do you control memory for one chat?
type /memories in the chatgpt desktop app or codex cli. it sets whether the current chat can use existing memories and whether codex can use the chat to generate future ones. openai says chat level choices don't change your global memory settings.
to wipe the store, select delete codex memories in the desktop app and confirm. openai says this resets the selected machine's memory store, and on the local machine it also clears computer history when that feature is available. we compare computer history with ReMynd in ReMynd vs chatgpt computer history.
what about chatgpt desktop mcp servers?
they live in the same codex config. openai's steps for the chatgpt desktop app: open settings, select mcp servers, select add server, enter a name, choose stdio or streamable http, give the server's command or url, then save and select restart. type /mcp in the composer to see connected servers.
codex stores that entry in config.toml, ~/.codex/config.toml by default. a codex config.toml mcp entry is a [mcp_servers.<name>] table, and a stdio server requires command, the command that starts it. once a server is configured you can switch among the desktop app, codex cli and ide extension without redoing setup. from the cli, codex mcp add does the same job. chatgpt web uses remote mcp tools through plugins instead.
where does ReMynd fit?
ReMynd is a desktop app that records your screen, runs ocr locally with apple vision on a mac, and keeps a searchable index on your computer. text comes from the focused window only, and video covers windows you have not excluded.
in our view, codex memories hold what you did in codex chats, and ReMynd holds what you saw on screen. ReMynd's Codex & ChatGPT switch sets up Codex on Mac and Windows (CLI, skill and the remynd-mcp entry in config.toml), which the ChatGPT desktop app shares; the ChatGPT website does not. connected agents get read-only history access and send what they read to their own cloud.
on a mac, ReMynd needs macos 14 or later on an apple silicon mac. ReMynd also runs on windows 11 (x64 or arm64). for the setup itself, see codex context from your screen history, and for claude, does claude desktop remember?.
one honest limit: ReMynd only has what was on screen while it was running, and ReMynd lets you exclude apps and sites you do not want in your history.
limits
menu names, defaults and file paths come from openai's chatgpt learn docs and the openai help center as read on november 20, 2026. openai notes that memory features and controls can vary by plan, region, platform and workspace settings, so yours may differ. this page does not cover chatgpt work, which openai says does not use a local codex memory store.
download the ReMynd desktop app and search what you saw in a focused window, by app and date.