Product Update
Memory Decay: How AI Remembers Like Humans Do
Today we're introducing Memory Decay — a system that makes AI memory behave more like human memory. Frequently used memories stay strong, while unused ones naturally fade. The result? More relevant, contextual AI interactions.
Marius Ndini
Founder · Jan 15, 2026
The Problem with Perfect Memory
Most AI memory systems treat all memories equally. A preference you mentioned two years ago has the same weight as something you said yesterday. But that's not how human memory works — and it's not how AI memory should work either.
When your AI assistant remembers that you "loved the Thai place on 5th Street" from 18 months ago with the same confidence as your current dietary restrictions, it creates awkward, irrelevant suggestions. The AI needs to understand that some memories matter more than others — and that relevance changes over time.
How Memory Decay Works
Memory Decay introduces a natural forgetting curve inspired by cognitive science. Here's what changes:
Confidence Scoring
When the AI extracts a memory from conversation, it assigns a confidence score based on how certain it is about the information. "User explicitly stated they're vegetarian" gets high confidence. "User might prefer morning meetings" gets lower confidence. Higher confidence memories are prioritized during recall.
Reinforcement
Every time a memory is recalled and used in a conversation, it gets reinforced — like how studying strengthens neural pathways. Memories that are frequently relevant stay strong. Memories that are never recalled gradually fade in importance. This creates a natural "survival of the fittest" for your AI's knowledge.
Temporal Awareness
Some memories are inherently time-sensitive. "User is traveling to Tokyo next week" should decay faster than "User's favorite color is blue." The system automatically identifies temporal memories and applies faster decay rates, so outdated information doesn't clutter your AI's context.
Source Tracking
Memories now track their origin: explicit (created via API), inferred (extracted from conversation), or corrected (user corrected an inference). This helps you understand where knowledge came from and how reliable it is.
What This Means for Your AI
With Memory Decay, your AI applications become smarter about what to remember and when. The benefits:
- More relevant responses — Active, reinforced memories surface first
- Less noise — Outdated information naturally fades away
- Better context windows — Limited token budgets are spent on what matters
- Human-like interactions — AI that remembers the way people do
Available Now
Memory Decay is automatically enabled for all Mnexium projects. You can view decay information — including confidence scores, reinforcement counts, and temporal flags — in the memory detail view on your dashboard.
No API changes required. Your existing memories will begin decaying naturally, and new memories will be created with full decay metadata. The system handles everything automatically.
Get Started
Memory Decay is included in all plans, including the free tier. Sign up at mnexium.com/dashboard to start building AI applications with human-like memory.