Why Codex history disappears after switching providers (and how I fixed it)

#productivity

I ran into an annoying Codex problem while switching between official OpenAI auth and relay/custom providers.

The sessions were not actually gone. What broke was session visibility.

In practice, the symptoms looked like this:

old sessions were visible under one provider, then disappeared after switching to another one
codex resume and Codex App did not always show the same history
changing only the rollout files was not enough

Why this happens

Codex stores provider-related session metadata in two places:

rollout files under ~/.codex/sessions and ~/.codex/archived_sessions
SQLite state in ~/.codex/state_5.sqlite

If only one layer is updated, the history may still look inconsistent. That is why a session can appear to be "missing" even though the actual data is still there.

What I built

I built codex-provider-sync to align both layers together.

It can:

inspect the current provider distribution
sync rollout metadata and SQLite state to a target provider
optionally update root model_provider in config.toml
create a backup before each run
restore from backup if needed
skip locked live rollout files and report them clearly

Windows GUI

I also added a Windows GUI so normal users do not need Node or npm anymore.

The GUI can:

scan the current .codex home
show provider distribution from rollout files and SQLite
let you pick a target provider and execute the sync
optionally update root model_provider
restore from backup
automatically keep the newest 5 managed backups by default, with configurable retention