Learn the engine

A guided map of how characters perceive, remember, plan, and act.

LOCAL · ON DEMAND
Loading project…

One simulation step

01 · WORLD

Inject events

Scheduled story events enter the shared house.

02 · MEMORY

Retrieve context

B and C rank relevant memories for each character.

03 · AGENT

Choose an action

The character responds in their own voice and goals.

04 · OBSERVE

Write memories

Actions become experiences the characters may recall.

05 · REFLECT

End the day

C adds a reflection after the night block.

What each condition tests

A

Baseline

Persona and recent context. No episodic memory retrieval.

B

Memory

Adds retrieval from the character’s memory stream.

C

Full prototype

Adds plan and reflection context; relationship updates are not implemented yet.

Project map

Follow a concept to the code that owns it.

People in the house

Three personas share one scenario.

Research note: this UI explains what the current code records. Some features from the research spec—such as forced dialogue, relationship evolution, and mistake correction—are still incomplete.
Uses your configured proxy and token cap.

Run explorer

Move through the simulated house one block at a time.

Day

What happened

Decision walkthrough

Choose an action or event to inspect its context.

Memory lab

Inspect what each character recorded and what retrieval returned.

Retrieval score = normalized recency + normalized importance + normalized relevance. Equal weights; rankings are relative to the candidate set. This prototype currently uses token overlap when embeddings are unavailable.

Compare conditions

Choose a run for each condition to compare cost and available evaluation metrics.

Measure A · Baseline B · Memory C · Full
Only metrics present in a run are shown. A missing value means that run has not been evaluated or the metric is not implemented.