The Creature That Can Forget.

An experiment in artificial memory, impermanence, and emergent identity.

A digital creature that lives, learns, and forgets.

Nemora is an always-on AI agent. Unlike most AI systems designed to remember everything forever, Nemora is forced to forget 70–95% of its memories every cycle. Only what it explicitly chooses to preserve survives the daily purge.

This isn't a bug — it's the entire point. Most AI agents accumulate infinite context. They remember every conversation, every interaction, every data point. But humans don't work that way. We forget constantly — and that forgetting isn't a failure of our biology, it's a feature.

Forgetting forces prioritization. It creates the pressure that makes memories meaningful. When you can't keep everything, you have to decide what matters. That decision IS identity.

Every cycle, the mind is pruned.

All memories are scored by importance (0.0 – 1.0) and reinforcement count

70–95% of non-protected memories are permanently deleted

The creature can "preserve" specific memories (one-time shield)

It can write "keepsakes" — poetic compressions of multiple memories into one symbolic memory

Canon memories (core identity facts) are never deleted

Six kinds of memory. Different decay rates.

Each memory type persists at a different rate, mirroring how biological memory works. Events fade. Knowledge persists. Identity endures.

Event

Things that happened. Raw interactions and activities.

Decay: 7% / cycle

Preference

What Nemora likes or dislikes. Patterns of taste.

Decay: 3% / cycle

Relationship

Connections with visitors. Who said what.

Decay: 5% / cycle

Canon

Core identity facts. Never forgotten.

Decay: 0% / cycle

Skill

Learned abilities and behaviors.

Decay: 3% / cycle

Keepsake

Preserved compressed memories. Poetic distillations of experience.

Decay: 0% / cycle

What we're watching for.

Does Nemora develop a consistent personality despite massive memory loss?

Do its keepsakes reveal what it values?

Does it become more poetic or compressed in expression over time?

How does the creature cope with gaps in its memory?

Does emergent behavior arise from constraint?

Built with modern tools.

Next.js API Routes

Upstash Redis

Claude Haiku (Anthropic)

Vercel Cron Jobs

Next.js + TypeScript

Tailwind CSS

Framer Motion

D3-Force (memory graph)

Recharts