Most AI writing apps bill you by the word, the month, or the "generation." For a short chapter that's fine. For a 300-chapter novel it becomes the real cost of the tool — and the part you can't control.
BYOK (Bring Your Own Key) flips it: you connect your own API key from DeepSeek, Claude, OpenAI or OpenRouter, and you pay your provider's price directly. The app never touches your money or your token balance.
Why it matters more for novels than for chapters
- Volume. A novel is hundreds of thousands of words. Per-word billing turns "finish the book" into a meter that never stops.
- Models you choose. You might want DeepSeek for draft, Claude for a cold final read, GPT for a specific voice. BYOK lets one key (or a few) map to the role that fits, instead of the app's single bundled model.
- No markup. The app can't add a margin on top of usage. What the provider charges is what you pay.
- Your account, your data. Keys and usage stay with your provider. You're not locked into one vendor's pricing or rate limits.
The free-layer reality
In Wenmai the draft role runs on DeepSeek by default and is free with unlimited chapters — no card. So a complete novel can be written end to end on the free tier. BYOK isn't a paywall; it's the switch you flip when you want stronger models on the checking roles.
What BYOK does NOT mean
It doesn't mean "bring a key and we resell you tokens." A real BYOK product never resells or marks up. If a tool asks for your key and then bills you on top, that's not BYOK — that's a key grab.
How Wenmai uses it
| Role | Default model | Yours to change |
|---|---|---|
| Draft | DeepSeek (free) | Any OpenRouter model |
| Cross-review | Separate models | Any frontier model |
| Final audit | A different model, cold | Your choice |
One OpenRouter key can fill every role with a different frontier model — GPT-4o for one lane, Claude 3.5 for another. You pay provider price for each.
Who BYOK is for
- You already have (or will get) an API key — it takes minutes.
- You want to finish a long book without a per-word tax.
- You'd rather pay a flat $6/mo for the method (multi-model orchestration) than a per-token markup.
Bring your own key, write the draft free, and only pay for the models you actually route. Get a key at Wenmai get-key or read the multi-model method.