Who it is forA developer testing the progressive retrieval runtime who sees worse recall results and wants the previous behavior back quickly.
The problemExperimental retrieval can misbehave in a real project, and switching it off should not require deleting memory data.
The outcomeProgressive recall is disabled through the rollback command, and memory data remains in place for the legacy path.
What you need first
- Python 3.9+ and Git installed
- A local Codex installation for the full install (the demo needs no Codex account or API key)
- Review access to your Codex hooks configuration, since the installer does not edit hooks.json
Steps
git clone https://github.com/libenxier-beep/codex-memories.gitcd codex-memoriespython3 scripts/agent_memory.py --helppython3 scripts/agent_memory.py --progressive-state-dir /absolute/owner-only/state progressive enable(inferred, check the README)python3 scripts/agent_memory.py --progressive-state-dir /absolute/owner-only/state progressive rollback(inferred, check the README)
Prompt for your agent
In the codex-memories repo, run python3 scripts/agent_memory.py --progressive-state-dir <owner-only-state-dir> progressive rollback. Report the command output and confirm the memory authority under ~/.codex/memories still exists and was not deleted.
Caveat. The README reports that a compact hidden seal scored 3 of 6 on Recall@5 and was rejected, so progressive retrieval is not yet shown to beat simpler recall in general.
Use this in your AI agent
Ask your agent
Using the Deduped MCP, get the setup steps for libenxier-beep/codex-memories and "Roll back progressive recall without losing memory"
Add Deduped to Claude Code once
claude mcp add --transport http deduped https://dedupedai.com/mcp
Every Monday: the 5 hottest repos and exactly how to use them. Free, one email a week, unsubscribe anytime.
Get them freeShare: X LinkedIn Hacker News Email Copy link
Third-party repos we don't own. AI-written use cases can be wrong. Disclaimer