ZenBrain Playground

The neuroscience-inspired, 7-layer memory architecture for AI agents.

Your AI forgets everything after every conversation. ZenBrain fixes that — with the same mechanisms your brain uses: spaced repetition, Hebbian strengthening, sleep consolidation and exponential forgetting curves. Not a vector database with a wrapper. Actual neuroscience.

The Memory Architecture

7 neuroscience-grounded layers · an overview map

Most AI memory is a flat store. ZenBrain is a 7-layer architecture shown as a depth space: permanence is depth, confidence is luminance, decay is cooling light, provenance a pullable thread. This map is illustrative — the panels below run the real numbers live.

↑ volatile · nearpersistent · deep ↓
input · document opened
fact · customer X · open quote
inference · follow-up due
Workingactive now · volatile
current task · high confidence
note · uncertain
open document
Short-Termthis session
conversation context
fades in 2h · decay
customer X mentioned
Episodicevents · time
meeting Tue · quote discussed
older episode · cooling
quote Q3 opened
Semanticfacts · knowledge
customer X · manufacturing
prefers morning callbacks
rarely-used fact · cooling
Proceduralworkflows · skills
quoting workflow
approval process learned
Cross-Contextlinked across contexts
linked: customer ↔ project ↔ team
strong Hebbian link
Corepinned · never decays
identity: „Sales DACH"
guideline: always valid
pinned by the user
Depth = permanence

Volatile flickers near the top; pinned identity rests deep and solid. You dive through the layers.

Luminance = confidence

Certain glows solid; uncertain is translucent and grainy. Confidence is seen, not read as a percent.

Warmth = freshness

Fresh glows warm; decaying cools and fades toward its countdown. Consolidation warms it back up.

Thread = provenance

„Why does the AI know this?" — a thread from input through fact to inference, pullable both ways.

Every number below is computed live by the real, open library — npm i @zensation/algorithms — running in your browser. Nothing is faked.

The Forgetting Curve

Ebbinghaus (1885) · FSRS spaced repetition

A memory decays as R = e−t/S. Turn on spaced repetition and watch each review reset the curve — so retention stays high with a fraction of the storage.

Spaced repetition — review on days 3 · 8 · 16
With spaced repetition
Without any review

Hebbian Learning

“Neurons that fire together, wire together”

When two memories are recalled together, the edge between them strengthens (asymptotically toward a cap). Left idle, it decays — and is pruned when too weak.

Edge weight
1.00
Co-activations
0
Status
neutral

Sleep Consolidation

Stickgold & Walker (2013) · memory replay

During sleep the hippocampus replays recent memories — important and emotional ones are strengthened, weak connections are pruned. Press sleep and watch it happen.

Replays this cycle
Weak edges pruned

Calibrated Confidence

Bayesian confidence intervals

ZenBrain never returns a bare number. Every estimate carries a 95% confidence interval that narrows as evidence accumulates — few reviews mean honest, wide uncertainty.

Point estimate
95% confidence interval