AI Model Selection

MythTapestry offers optional AI-powered features to help you build and explore your world. No AI feature is required — you can use MythTapestry purely as a manual world-building tool. Some smart features like entity detection can run entirely locally without any external AI calls.

When you do choose to use AI features, they are powered by models available through OpenRouter, and you have full control over which model is used for each feature.

Per-feature models

Each AI feature can use a different model so you can balance cost, speed, and quality independently:

Feature	What it does	What kind of model to pick
Entity Detection	Spots references to your entities while you write	Fast, cheap — runs often
Entity Assist	Proposes field values for new or existing entities from your scenes	Mid-tier — reading + writing in one pass
Relationship Suggestions	Spots new or ended relationships in what you’ve written	Good at reasoning over short passages
AI Summaries	Writes a recap of a scene or chapter	Good at prose, doesn’t need to be huge
Chat	Conversation that answers questions about your world	Strong reasoning, large context window
Query rewriting	Resolves “she”, “they”, “the same one” in chat follow-ups	Fast, cheap — runs on every turn
Auto-title	Names new chat sessions automatically	Fast, cheap

Note

Entity detection also offers a GLiNER mode that runs entirely on the server without any external AI calls. See Entity Detection for details.

Configuring models

Model preferences belong to you, not to a world: the model you pick for a feature is used in every world you open. Set them from your account menu → Profile settings → AI Preferences. See Profile & AI Preferences.

Which model gets used

When a feature runs, MythTapestry picks a model in this order:

1. Your profile preference, if you’ve set one
2. The feature’s built-in default
3. The cheapest model we make available that can do the job

The first one that exists wins. You don’t need to set a preference for everything — leaving a feature unset just means it uses the built-in default.

Consistency checks bend step 2, because the checks in that one feature don’t all need the same model. Checking whether Aldric could make a journey is a light check, so it uses the cheapest model we include for consistency checks. Checking whether he knows a secret, and reviewing a whole chapter, need close reading, so they use the model we’ve picked for careful work. Checking a single action uses the ordinary default. Your own preference for consistency checks still wins over all of these.

Tip

Bring your own key: if there’s a model you want to use that isn’t on our included list (a specific OpenAI, Anthropic, or DeepSeek model, say), you can paste an OpenRouter API key into your profile settings — on plans that include bring-your-own-key (Mythwright). From then on that model is selectable for any feature, and the usage is billed directly to your key — it doesn’t count against your MythTapestry quota.

Usage quotas

Each AI feature has its own monthly allowance that resets on the same day every month. You can see how much you’ve used on the Profile page under Usage & Billing, broken down per feature. If you run out for a feature, that feature simply pauses until the reset — other features keep working.

One click is one use, however much work the AI does behind it. Suggest Updates in Entity Assist, for example, may read and summarise dozens of scenes before it answers, and that still counts as a single Entity Assist use.

You aren’t charged when the AI’s answer comes back in a form MythTapestry can’t read, even if you still see a fallback (a summary made by trimming the scene’s text, say, or a consistency check made from simple rules). An answer that reads fine but finds nothing — no new characters in a scene, no updates for an entity — still counts, because the AI did the work.