Hey everyone! I'm the maker of Mistikguard. I built it while developing my own local AI companion because I kept running into the same frustrating problem: the model would occasionally "remember" things that were never actually said, and then confidently repeat them later. Mistikguard is a lightweight Python library that tries to reduce memory fabrication by: - Distinguishing between facts the **user confirmed** vs what the model just **inferred** - Using a write gate to block contradictions and self-narration - Making user corrections more durable with "tombstones" - Adding an optional grounding audit after responses It's designed to be simple to integrate and works with Ollama and other local models. This is still early (v0.1), so I'd love to hear your thoughts especially if you're building AI companions or agents. What kind of memory problems are you running into? Happy to answer any questions!

Mistikguard

Prevent memory hallucinations in your LLM applications

Prevent memory hallucinations in your LLM applications