An AI that remembers your whole story.
A novel has a canon — characters, timeline, voice, the rules of your world. Memrith keeps an editable story bible your AI draws on, scoped to the project and yours to keep — so chapter forty stays true to chapter two.
It knows the canon, not just the last thing you pasted.
Why long writing breaks AI chat
A manuscript is the one project guaranteed to outlast a chat.
Ask an AI for help with chapter two and it does fine. Ask it again at chapter forty and the cracks show. By then your protagonist has a backstory you settled in week three, a city you renamed in week eight, and a death you moved earlier in the timeline last month. The model can only hold so much at once. A single conversation runs out of context window long before a novel ends, so the early chapters fall out of view exactly when you need them to stay in — and it cheerfully gives a character the wrong eye colour, or has someone attend a wedding three chapters after you killed them.
ChatGPT and Claude both added memory across chats in 2025 and 2026, so this is no longer "the AI forgets everything." That part is real, and it helps. But the memory they keep is account-wide: it blends your novel with your day job, your short stories, the email you asked it to draft, the recipe you looked up. For a writer that blending is the problem. Your manuscript's canon isn't a vibe assembled from everything you've ever typed — it's a specific, deliberate set of facts for one book, and you need the AI working from that and nothing else.
There's a second gap. Account-wide memory is something the model decides to keep, in its own words, inside one company's product. It isn't a document you open and correct line by line, and it can't follow you to a different model. For a long project — the kind that takes years, where you'll change tools at least once — those two things matter more than they look.
What a story bible actually needs
Writers have kept story bibles since long before AI: a binder, a wiki, a spreadsheet of who's who. The reason is always the same — a long story holds together because of decisions you can't keep entirely in your head. For an AI to help without contradicting you, it needs that same bible, in a form it can use. Concretely:
- A canon, scoped to the project. The facts of this book — not blended with every other thing you've discussed. Chapter forty should be checked against chapter two, not against last week's grocery list.
- Character and timeline consistency. Each character's facts and voice; the order events happen in. The two places contradictions creep in first — eye colour, who knew what when, the body that won't stay dead.
- Your voice, not a generic one. How you write this world — terse or baroque, wry or grave. A bible that captures voice keeps the AI's suggestions sounding like your book instead of like everyone's.
- Editable, by you. Canon changes. You rename a city, cut a subplot, age a character up. You need to edit the record directly and have every later answer follow the correction — not argue with a memory you can't reach.
- Per-project, kept apart. The space opera and the quiet literary novel should not bleed into each other. Each project keeps its own canon.
How Memrith holds the canon
Memrith keeps your story bible as editable entries — characters, places, timeline, the rules of the world, your voice — that live in a workspace for that project. When you draft or brainstorm with your AI, it reads the relevant entries first and answers from the established canon, not from whatever happened to be in the last few messages. The bible survives closing the app, coming back next week, and starting a fresh chat. You don't re-paste a summary every session; the continuity is already there.
Because it's scoped per workspace, your two books don't blend. Open the fantasy project and the AI knows that world's geography and gods; open the thriller and it knows that timeline and cast — and neither one leaks into the other. And because the canon is an editable record rather than a guess, when you change something — rename the capital, kill a character three chapters earlier — you edit the entry once, and the change holds across every later session and every model.
One concrete example. You decide in revision that your antagonist was never at the harbour fire — she ordered it from the capital. You open her entry and the timeline entry, fix both, and lock the harbour fact. From then on, in this project, the AI stops placing her at the scene — next week, next chapter, even after you've switched from Claude to GPT. You corrected the canon once; you didn't re-explain it.
Built-in AI memory vs a story bible you own
To be fair about it: ChatGPT and Claude do have memory now, and for casual use it's good. The difference for a long manuscript isn't whether the AI can remember — it's whether the thing it remembers is a canon you control and can take with you.
For a casual question, account-wide memory is fine, and free. For a book you'll work on for years, the thing you want is a canon you curate and keep. That's the case Memrith is built for — and it's the same logic behind keeping any long project's memory scoped and owned, which is the broader idea on the AI project memory page.
Own it, take it anywhere
The story bible is a file on your machine, in a documented format you can read, edit, and export. That has two consequences writers tend to care about once the project gets long.
First, it's yours to keep. The canon you build over a year isn't a feature you rent inside someone's product — it's a file you own, never used to train a model, exportable as plain text or Markdown any time. If you stop using Memrith, the bible leaves with you. That's the whole argument of the own your memory page, applied to the thing you'd be most upset to lose: the canon of your book.
Second, it's portable across models. Halfway through a draft you might decide a different model is better for dialogue, or cheaper for the long grind of revision. With built-in memory that switch means leaving your accumulated context behind. With Memrith, switching from Claude to GPT to OpenRouter is a Settings change — your story bible is unchanged, and the next chapter opens with the same canon, routed through the new model. You bring your own key and pay the provider directly; the reasoning is on the why BYOK page.
The honest version. If you only ever ask an AI the occasional question about your story, the built-in memory in ChatGPT or Claude is simpler and free, and Memrith is more than you need. Memrith earns its keep on the long project — the manuscript with a real canon you'll revise for years and probably outlast a provider on. Setup is a few minutes (bring your own AI key), and that's the trade for a bible you own and can take anywhere.
Common questions
Can AI remember my novel's canon across chapters?
Not reliably on its own — a single chat runs out of context window long before a novel ends, so the model loses the early chapters as you write the later ones. ChatGPT and Claude now carry memory across chats, but it's account-wide and blends every project you've ever discussed, not a canon you control for one book. Memrith keeps an editable story bible scoped to that project — characters, timeline, the rules of your world — that your AI draws on every session, so chapter forty can stay consistent with chapter two.
Is there an AI for worldbuilding that remembers?
Memrith holds the canon of a world as editable entries scoped to the project's workspace: the map, the magic system's costs, who rules what, which names are taken. Because it's a file you own rather than a prompt you re-paste, the worldbuilding accumulates across months instead of resetting each session — and your AI answers from the established canon rather than inventing a new continent every time.
How do I keep AI consistent with my characters and timeline?
Write the canon down once, where the AI can use it. In Memrith you keep entries for each character's facts and voice and for the timeline of events; the AI reads them before it drafts, so it stops giving a character the wrong eye colour or putting a death before the wedding. When the canon changes — you rename a city, kill a character early — you edit the entry, and every later answer follows the correction.
Can I edit what the AI remembers about my story?
Yes — that's the core of it. Your story bible is an editable file you own, not a model's guess you can't reach. Open any entry to correct it, lock a fact you don't want drifting, merge duplicates, or delete a thread you cut. The canon is the source of truth, and you are the one who decides what it says.
Will my story bible move if I switch AI models?
Yes. The story bible lives on your machine, not inside one provider, so switching from Claude to GPT to OpenRouter mid-draft is a Settings change — the canon is unchanged and the next chapter opens with the same context. Built-in memory inside ChatGPT or Claude can't do this: ChatGPT's memory can't move to Claude, and Claude's can't move to GPT. With Memrith the model is replaceable; the bible stays.