Chatbots today remember facts, but they don’t reflect, and they never know when they’re blank. That’s why they repeat mistakes, serve stale knowledge, or confidently hallucinate.
In this lightning talk, I’ll show memori, a tiny FOSS Python library that fixes this with three simple ideas:
CRUD: add, get, update, and delete memories.
Reflection: clean up duplicates, flag stale info.
Gap detection: instead of faking, the agent asks, “Teach me.”
Live demo: a FastAPI + HTMX journaling bot where you edit memory, press “reflect,” and instantly watch the agent behave differently.
In 15 minutes, you’ll walk away knowing how to add reflective memory to your own apps with just Docker and a few hundred lines of Python.
Reflection + gap detection are the missing pieces in agent memory.
memori runs locally in minutes; you can try the demo bot yourself.
Sometimes the simplest FOSS building blocks solve the biggest reliability gaps.