BYOK and Local AI Models for Novel Writing: Cost, Control, and Privacy
Understand hosted credits, bring-your-own API keys, OpenAI-compatible endpoints, and local Ollama models when choosing an AI setup for fiction writing.
BYOK means you connect your own provider key or endpoint, while a hosted model uses the platform's managed provider. The right choice depends on who pays the model, where requests travel, which models you need, and how much setup you want to maintain.
Hosted models and BYOK are different trade-offs
| Setup | Who pays the model provider? | Main advantage | Main responsibility |
|---|---|---|---|
| Hosted model | The platform, through managed credits | Fastest setup | Track plan limits and available models |
| BYOK endpoint | You pay the provider directly | Choose provider and model | Protect the key and monitor provider billing |
| Local Ollama | Your local machine | Keep inference local and control the runtime | Install, update, and provide enough hardware |
Check the current pricing and provider terms before making a cost decision. Token prices, model access, and limits change; do not treat a blog post as a billing contract.
What to check before connecting a provider
- Confirm the endpoint is compatible with the interface the app supports.
- Use a key with the smallest practical permissions and a spending limit.
- Test a short request before sending a full manuscript context.
- Check which model name and context window the provider actually returns.
- Read the provider's retention and training terms for your account tier.
- Remove or rotate the key if you shared it in logs, screenshots, or support messages.
NovelKnow supports managed models plus custom providers compatible with OpenAI-style endpoints and local Ollama setups. The provider settings can test a connection and sync available models; the exact options depend on the current deployment.
Privacy is a data-flow question
Before sending a scene, identify:
What is sent: selected prose, relevant Codex entries, prompt instructions, and generation context.
Where it goes: the connected provider or local runtime.
Who controls retention: the provider's current policy and your account settings.
What remains in NovelKnow: project data, provider configuration, and usage records covered by the app's privacy policy.
Do not assume that “local model” means every part of the workflow is local, or that “BYOK” means a provider cannot retain requests. Verify the route and terms for the setup you choose.
A practical setup for a long project
Use a hosted model for a quick first session, a BYOK provider when you need a particular model or direct billing, and a local endpoint when your hardware and quality requirements make that worthwhile. Keep the manuscript's stable facts in the writing workspace, but send only the context needed for each request.
For sensitive material, test the workflow with invented names and a short passage first. Confirm the output, logs, provider dashboard, and deletion controls before using unpublished work.
Frequently asked questions
Does BYOK make AI usage free?
It avoids the platform's managed-credit charge, but the connected provider may charge for requests. Local inference also has hardware and electricity costs.
Is a local model always better for privacy?
It can reduce the path your manuscript takes, but only if the runtime and surrounding application are local and configured safely. Confirm the actual network route.
Should I send my entire novel to every request?
No. Send the scene goal, relevant facts, and necessary context. Smaller, intentional context is easier to review and reduces accidental disclosure.
Turn this guide into your next scene
Keep your outline, character notes, and draft together in NovelKnow. Start with one scene and review each AI suggestion before keeping it.