Best experienced on desktop

Wenmai's writing studio needs a large screen to configure your AI model keys and run multi-model collaborative writing. Mobile is fine for browsing features — open wenpath.com on a desktop browser to start writing.

Not another chat box — an AI editorial team

One model writes a great chapter.
A system finishes the book.

Bring 1 API key · finish a whole novel · no chapter cap

You've probably hit this with AI novel writing: the first chapters are great, then it falls apart — a character acts out of character, a planted clue is never paid off, the voice drifts. That's not a weak model. It's one AI doing both the writing and the editing. Wenmai splits the work: four AI roles each own one lane and cross-check. You bring your own key, the draft role is free and unlimited, and you can finish a whole book.

✓ Bring your own key — billed at provider price, never resold✓ Draft role free, unlimited chapters✓ Your key stays in your browser
Letters from the Fog — control deskFinal draft signed off
Final deliverables · Ch. 042—045

Ready to publish

This batch has passed canon verification, a cold read of consecutive events, historical-error matching, and platform publish checks.

No intermediate drafts, no alternatives, no line-by-line edit suggestions. The author only copies the text and publishes.

Delivery receipt 4 chapters · 9 canon checks · 18 regressions
Canon always onNo candidate draftsYou steer, not editClean pass to publish
How it works

The book doesn't collapse
because writing and checking are separate.

Swap any model or provider — this flow doesn't change. It's the reason a novel stays consistent past chapter forty, and past a few hundred chapters. You confirm the key nodes: outline, event package, character setup and the draft are all viewable, editable, or rejectable.

01

Canon

Align to the fact ledger

Before a word is drafted, the chapter aligns to the canon — characters, items, timeline, open foreshadowing. Contradictions are blocked at the source, not found after.

02

Draft

One model writes the chapter

A single dedicated draft role writes the prose. That's its only job. No editing, no self-review — just the chapter, well written.

03

Cross-review

Three roles check it

Continuity verifies facts. Character-voice guards each person's way of speaking. A different model does a cold final read. Writing and checking are separate lanes.

04

Final audit

A clean pass to publish

What comes back is a finished chapter with a delivery receipt — not a pile of 'here are some suggestions.' Copy and publish.

Comparison · Novelcrafter alternative

Looking for a Novelcrafter alternative?

Most novel tools keep a world bible and still leave one model to both write and edit. Wenmai is BYOK and splits the work across four roles. Here's the honest difference.

Single-model novel appWenmai
World bibleOften yesYes — a canon ledger
Who writesOne modelOne dedicated draft role
Who checks factsYou, by handA separate continuity role
Who guards voiceYou, by handA separate character-voice role
Final readYouA different model, cold
API keyOften the app's, billed by wordYours (BYOK), at provider price
Free to finish a bookDependsYes — draft role, unlimited chapters

The point isn't "more models." It's that a system organizes them. Without that system, four models is just four copy-paste tabs. Read the full comparison in our guide: Novelcrafter alternative: what to look for.

Bring your key, it's a full tool

Bring your key and write.
It finishes a whole book.

Wenmai doesn't resell models or take a token margin — you use your own API key. The $6/mo plan supports the multi-model orchestration method's continued development; it unlocks the cross-review roles, not any token or quota.

Free — bring your key

Draft role freeunlimited chapters, no time limit

The draft role (DeepSeek by default) has no chapter cap — enough to finish a whole novel. Canon ledger, event-package design, and draft creation are all on.

Start writing free
  • Canon ledger, event-package design, draft creation — all on
  • Novel / short-drama / review / continue modes, unlimited
  • No chapter cap; long-form memory and canon run as normal
  • Bring your own API key, billed at your provider's price

Payments are handled by Creem (merchant of record). Wenmai is a complete writing tool on its own; the $6/mo plan supports the multi-model method's development and unlocks the cross-review roles.

FAQ

Common questions

Do I have to pay to finish a novel?
No. The draft role is free and has no chapter cap — bring your own DeepSeek (or any) key and write from chapter one to the ending. Wenmai never resells tokens or marks up your usage; you pay your provider's price. The $6/mo plan unlocks the cross-review roles and mistake regression, which make a long book more consistent. They're optional, not a gate.
What does BYOK mean, exactly?
Bring Your Own Key. You create an API key at your provider (DeepSeek, Claude, OpenAI, Kimi, and others), paste it into Wenmai, and usage is billed at your provider's price. Wenmai holds the key in your browser and never resells access or adds a margin.
Is this a Novelcrafter alternative?
It covers the same job — keeping a long book consistent — but differently. Most tools keep a world bible and still leave one model to both write and edit. Wenmai splits writing from checking across four roles, and it's BYOK. If you already have a key and a book longer than thirty chapters, the consistency gain is the reason to switch. See the full comparison in our guide.
Why not just use a stronger single model?
A stronger model still has one context window and one set of eyes. By late chapters it can't hold the whole book, and it reads its own draft with the same bias that wrote it. The fix isn't a smarter model — it's separating writing from checking. That's what the four roles do.
Will the writing sound like AI?
It depends on how you use it. If you let one model dump the whole book, yes. Wenmai's design fights this: the canon pins character and foreshadowing, the event package gives each chapter a clear goal, and the character-voice role checks that people sound like themselves. A human pass at the end is still recommended — but the raw output is far closer than single-model straight-through.
Can I bring my own Claude or OpenAI key?
Yes. Any role can use any model you connect. The only guideline: the final-audit role should use a different model than the draft role, so the cold read is actually cold. You choose the providers and model IDs.