Show HN: Codex-mem, 90% fewer tokens for Codex

1 points by alfredray 8 hours ago · 0 comments · 1 min read

I built codex-mem to give Codex persistent memory across sessions.

Benchmarked result: - 99.84% token reduction (379,275 -> 596) - ~60ms median to first memory context

It uses progressive retrieval (search -> timeline -> details), plus MCP tools and local-first privacy. Repo: https://github.com/StartripAI/codex-mem

No comments yet.