projectmem v0.1.4 is live — the accountable-judgment release

Hey everyone 👋 Three weeks after launch, projectmem v0.1.4 is out — the release I'm most proud of.

The core idea got sharper: your AI's memory now flags its own staleness instead of silently trusting (or deleting) old decisions.

What's new:

✅ Stale-memory detection — when a decision's file has moved on in git, projectmem flags it ("predates 7 commits — confirm or supersede"). It never deletes; you decide.

✅ Supersede — retire an old decision without losing history (the log stays append-only, always)

✅ pjm brief — a one-screen "where was I?" briefing for the start of every session

✅ Pre-commit warnings now list what already failed, so you never re-try a known dead-end

✅ pjm export --claude-md — even agents without MCP inherit your project's judgment

✅ New dashboard Overview (screenshot 👇)

Still 100% local. No daemon, no telemetry, no cloud account. MIT.

There's also a research paper behind it now (arXiv:2606.12329) and a full tutorial video.

pip install -U projectmem

Would love your take 🙏 — what would make your AI's memory feel more trustworthy?

📰 https://projectmem.dev/blog/projectmem-v014-stale-memory-detection-ai-coding-agents

🎓 https://arxiv.org/abs/2606.12329

🎬 https://youtu.be/pELGdXHj_Ls