Feature request: manual Codex memory refresh and optional shared memory with ChatGPT

Summary

Codex memory currently appears to rely mainly on automatic/background triggers. In practice, this makes the memory feature feel less controllable than ChatGPT memory: users can ask ChatGPT to remember or update something during a conversation, while Codex users do not seem to have an obvious manual action to refresh, consolidate, or apply memory updates on demand.

Pain points

• Memory generation/consolidation is automatic and not directly user-triggerable.
• The trigger conditions feel strict and opaque, for example idle-time windows, rollout age windows, startup batching, or rate-limit conditions.
• When a user writes an ad-hoc memory update note, it is not clear when it will be absorbed into MEMORY.md / memory_summary.md, or how to request immediate consolidation.
• Older threads may fall outside the default automatic memory window, even though they can contain important long-term project context.
• Compared with ChatGPT memory, Codex memory feels harder to inspect, correct, and update in the moment.

Requested improvements

1. Add a manual memory refresh/consolidation action in Codex.

• Example: a command, button, slash command, or settings action such as “Refresh memories now”.
• It should show whether new notes, recent threads, archived sessions, or rollout summaries were considered.

1. Add clearer memory status and diagnostics.

• Show last memory update time.
• Show pending ad-hoc notes.
• Show why memory generation was skipped, such as idle time, age window, or rate-limit threshold.

1. Make ad-hoc memory notes easier to apply.

• If a user explicitly asks Codex to write a memory update note, there should be an obvious way to trigger consolidation afterward.
• Ideally Codex could report: “this note has been incorporated” or “this note is pending”.

1. Consider an optional shared-memory switch between ChatGPT and Codex.

• Many users use ChatGPT for high-level thinking and Codex for local execution.
• It would be useful to optionally share selected long-term user/project memories between ChatGPT and Codex.
• This should be opt-in, inspectable, and controllable, because personal memory and project execution memory have different privacy and relevance needs.

1. If OpenAI is planning a future unified “super app” experience, consider making memory interoperability part of that design.

• ChatGPT could keep broad personal/context memory.
• Codex could keep project/workflow/execution memory.
• Users could choose what is shared, imported, or kept separate.

Why this matters

Codex is often used across long-running local projects. Important context can live in older archived threads, project-specific workflows, and user-written memory notes. If memory only updates automatically under strict conditions, users cannot reliably correct or enrich Codex’s long-term context when they notice something missing.

A manual refresh plus transparent diagnostics would make Codex memory more predictable and closer to the usability of ChatGPT memory, while still preserving Codex’s local/project-oriented memory model.