codex-memory-manager

To persist user preferences from Codex conversations, you can automate durable memory management by extracting the last 24 hours of session data and writing categorized memory files. This captures stable preferences without requiring manual re-entry of transient instructions.

Automated memory management for Codex works by reading session stores at ~/.codex/sessions and ~/.codex/archived_sessions, distilling stable user preferences, and writing categorized memory files to ~/.codex/memory/*.md. It then updates the AGENTS memory index for future recall.

The best way to organize Codex memory files is by categorizing them into structured markdown files under ~/.codex/memory/*.md. This normalized index speeds up future agent responses by maintaining stable preferences separately from transient session instructions.

You do not need extensive configuration to extract preferences from the last 24 hours of Codex sessions. The process reads existing session stores at ~/.codex/sessions and ~/.codex/archived_sessions directly, requiring no additional dependencies to begin capturing and storing memory.

Yes, you can update the AGENTS.md memory index automatically from recent chat history. A dedicated synchronization script rewrites the end of ~/.codex/AGENTS.md to reflect newly extracted or updated memory categories from your latest Codex interactions.

Your Codex agent forgets preferences because session instructions are transient and not stored durably. By automating memory extraction from recent conversations and updating the memory index, the agent gains consistent recall of your stable preferences across sessions.