@embirico@OpenAIDevs@sama My Codex App automatically deleted almost all of my threads today.
I’m a GPT Pro subscriber, and this was a genuinely scary data-loss incident.
The threads were not in Archive. The app still had thread metadata in SQLite, but the underlying rollout JSONL files were gone/unlinked from ~/.codex/sessions and ~/.codex/archived_sessions.
After digging locally, this did not look like a normal “archive” or “delete thread” operation. The DB still referenced the missing rollout paths. Some deleted rollout files were even still held open by the Codex app-server process with link count 0.
The only way I could make the thread list usable again was to create placeholder rollout files, which restores the thread entries but not the original transcripts.
For anyone relying on Codex as an agent workspace: this is exactly the kind of failure mode that makes people afraid to trust long-running AI work. Threads are not just chat history; they contain project state, decisions, debugging context, implementation trails, and handoff records. Losing them silently is a serious reliability issue.
Please treat this as a P0 data-loss class bug:
1. Investigate why rollout files can be cleaned/unlinked while DB thread records remain.
2. Add integrity checks before any cleanup/migration touches user session files.
3. Provide a real recovery/export/backup path for Codex App users.
4. Warn users if local session storage is at risk.
I’m posting this because I don’t want other users to find out the hard way that their agent history can disappear overnight.
@embirico@OpenAIDevs
Codex grew programmatic policies with no neural nets: max score on Breakout, and SOTA-level scores on MuJoCo.
Maybe heuristics were not too weak. Maybe they were just too expensive to maintain. Maybe it's the next paradigm.
https://t.co/1ZaIneleuW