OpenAI Codex Cloud agents gain persistent workspace memory across sessions

Read the original article →

What happened

OpenAI added persistent workspace memory to Codex Cloud agents on July 19. Long-running coding agents can retain repository context, prior decisions and open tasks across sessions without re-ingesting the full codebase each time.

Memory is scoped per workspace with admin controls for retention and deletion. The feature is available to Codex Cloud customers on GPT-5.6 Terra and Luna tiers; Sol remains preview-only. OpenAI said memory reduced repeated clarification loops by roughly 30 percent in early testing.

Why it matters

Agent reliability often breaks when context resets between sessions. Persistent memory targets a practical pain point for teams running multi-day refactors or migration projects.

Admin retention controls matter because memory can accumulate sensitive code and decisions that should not persist indefinitely.

MintedBrain take

Memory can also preserve mistakes. Set retention policies and periodically clear workspace memory on long projects, especially when agents change direction or when repositories contain secrets that should not linger in agent state.

References

This article was originally published at The Verge. For the full piece, read the original article.

Discussion

  • Loading…

← Back to News