Narrative Memory: the AI that knows your novel
Narraya's narrative memory lets the AI know your book's characters, places and tones. Isolated per novel, respectful of privacy.
Anyone who has tried asking ChatGPT for advice on a chapter of their own novel knows the same ritual: paste in a synopsis, recall who the protagonist is, explain the tone, describe the context. Every session. As if the reader changed each time. It is the most frustrating limitation of generic AI models: they don't remember your book, they don't recognize it, they don't inhabit it.
Narraya's narrative memory exists to solve exactly that.
What a narrative memory is
Every book you create in Narraya builds, progressively, a structured "memory" of itself. Characters, places, key events, dominant tones, description styles, narrative arcs as they emerge. It isn't a summary someone writes for you: it is a map that grows with your novel, updated every time you close a chapter, fill in a character sheet, or map a relationship.
When you then query Narraya — for a chapter analysis, a writing feedback, a coach suggestion — the AI doesn't start from zero. It already knows the novel is a noir set in Trieste, it already knows Elena is thirty and coming out of a failed marriage, it already knows the first body appeared in chapter three. What you ask arrives already framed.
Generic AI vs AI with narrative memory
| Aspect | Generic AI (ChatGPT, Claude direct) | Narraya with narrative memory |
|---|---|---|
| Character knowledge | None; you rebuild it each time | Persistent and consistent per book |
| Chapter-to-chapter consistency | Not automatically detectable | Flagged where missing |
| Text privacy | Depends on the service, often unclear | Content never enters public training |
| Cost per request | Flat but without context value | In credits, with context already loaded |
| Initial setup | Every session, on you | None: the memory builds itself |
One book, one memory
Something we care deeply about: the memory is isolated per book. If you're writing a fantasy set in an invented world and, in parallel, a realistic noir set in the Friuli region, the two memories don't talk to each other. The "Anna" in your noir doesn't contaminate the "Anna" you imagined as a twilight elf. It is a technical guarantee — the AI receives only the context of the book you're working on — but also a narrative one: every novel deserves its own bubble.
Your book's content — chapters, sheets, dictionary — lives in our databases and is never used to train public AI models. Your novel is yours: the narrative memory is a working tool, not a dataset.
What it remembers, in practice
A noir in Trieste
Protagonists, suspects, crime locations, timeline of events, the dark tone of night-time descriptions. By chapter eight you already know whether the inspector's alibi holds.
A multi-volume fantasy
Races, invented languages, political maps, magic rules. The book dictionary integrates with the memory: "Athelgar" isn't flagged as a typo every time.
A family saga
Four generations, twenty characters, family trees, dates, nicknames. The memory holds the puzzle together when you lose a detail in chapter fourteen.
A memoir
Real places, real people, historical periods. The memory helps you maintain consistency on dates and biographical details — precious when the past is eighty years old.
How it builds
You don't have to do anything special. Narrative memory updates in two ways: automatically, by reading the chapters you write, and manually, when you fill in character sheets, the book dictionary, and relationships between characters. The more structure you add, the sharper the memory becomes. But even without explicit sheets, the memory works: from the chapters it extracts names, places, tones, recurring patterns.
Technically, behind the scenes, we use pgvector — a semantic search technology that lets us retrieve the most relevant memory fragment for each request — along with structured metadata. But it's nothing you need to think about: the complexity sits under the hood, you're left with the writing.
Writing a long novel is, in large part, a work of memory. Narraya doesn't replace it — no AI can remember the things you meant to say — but it takes the weight of the mechanical details off your shoulders. The result is that, when you sit down to write, you have more mental room for what actually matters.
Curious about how the memory appears inside the product? The live demo shows, on a sample book, the kind of context Narraya passes to the AI on every request.