Mnemo - Auditable, citation-backed memory for production AI agents

I built Mnemo because I kept running into the same problem — agent memory tools that couldn't handle change. A user updates their job. Moves cities. Says "actually I'm not vegetarian anymore." And the agent just... keeps answering with the old fact. Not because it forgot. Because it never knew the old fact was old. Most tools store the latest value and call it done. That's not memory. That's a variable. Mnemo tracks how facts change over time bi-temporal versioning on an event-sourced core. So when someone says "she left that job last month," it handles that correctly. That means agents can answer "what's true now?" and "what was true six months ago?" from the same memory store. I didn't want to just claim it works. So I benchmarked it. 85.2% on LongMemEval-S and also made the full methodology public: https://github.com/shhahhussain/... If you've built an agent that has to remember someone across sessions — I'd genuinely like to know where it breaks for you. What's the memory failure that's been hardest to fix?