Skip to content

How the AI Remembers

AI models have no memory between requests. Every time you generate a scene, the AI starts from scratch. Storywright solves this by building a rich context for every scene — so scene 12 knows what happened in scenes 1–11, even when the full text doesn't fit.

Open Story Context in the left panel to see the AI's understanding of your story through three tabs.

Events — What Happened

The Events tab shows memories extracted from each scene — topic-tagged facts like:

  • "Elena discovered the hidden passage behind the bookshelf"
  • "The council voted to close the border"
  • "The forest glows blue at night due to bioluminescent fungi"

Before generating a new scene, the app selects the most relevant memories and includes them in the prompt. This way the AI recalls what matters without needing the full text of every previous scene.

Tip: Enable semantic embeddings in Settings → Memory for better recall accuracy. The app can run embeddings locally in your browser — no API calls needed.

World — What's Currently True

The World tab shows the current state of every tracked entity — characters, objects, locations — as readable cards:

  • Descriptions — physical traits, injuries
  • Status — situational state ("wanted by the guard")
  • Possessions — what they carry
  • Relationships — how entities relate to each other
  • Quantities — numerical facts ("40 days of fuel remaining")
  • Death — permanently marked when a character dies

This state is built from events extracted from each scene ("Elena gained the Moonblade in scene 3", "Marcus died in scene 7") and injected as constraints so the AI doesn't contradict established facts.

Fixing mistakes

Click Manage events at the bottom of the World tab to see every individual event. If the AI extracted something wrong, delete it — the world state recalculates immediately.

Issues — Quality Checks

The Issues tab surfaces two types of problems:

  • Lorebook conflicts — click Scan for conflicts to check for contradictions between entries
  • Story review findings — after running a Story Review, results appear here with severity, scene references, and suggestions

This is your quality checkpoint — a quick way to spot contradictions before they compound.

Behind the Scenes

In addition to Events and World state, the app generates scene summaries — 2–3 sentence recaps of each scene that give the AI the big-picture narrative arc. Together, these three systems cover different needs: summaries provide the story arc, memories provide specific facts, and world state prevents logical errors.

All of this extracts automatically after every generation and revision. You don't need to manage it — but you can review and correct it anytime through Story Context.

Tips

  • Check the World tab when the AI gets things wrong. Wrong location? Forgotten injury? The extraction may have missed something. Use Manage events to fix it.
  • Check Issues after batch generation to catch problems before doing detailed revisions.
  • Use lorebooks for critical facts. Memory recall is probabilistic — if something must always be included (a magic system rule, a character's backstory), add it as a lorebook entry with keywords.