Relationship Suggestions

MythTapestry can optionally analyse your content and suggest new relationships between entities — or detect when existing relationships have changed. This feature is off by default and must be enabled in World Settings.

How it works

When you save content that mentions multiple entities, the AI can:

1. Detect new relationships — “Aldric serves the Crown” suggests a “serves” relationship between Aldric and The Crown
2. Detect ended relationships — “Aldric left the Crown’s service” suggests the “serves” relationship has ended

Each suggestion includes the evidence text — the snippet from your writing that triggered it.

Suggestion types

Type	What it means	Action on confirm
New	A relationship that doesn’t exist yet	Creates the relationship in the knowledge graph
End	An existing relationship that has ended	Sets valid_to on the existing relationship

For “end” suggestions, the system links to the existing relationship it thinks has ended, so you can verify before confirming.

Triggering suggestions

Relationship detection runs in two ways:

On save (automatic)

When you save a content unit that has confirmed entity mentions and the feature is enabled, detection runs automatically on the full document.

Requirements:

• The feature is enabled in World Settings
• The content unit has at least 2 confirmed entity mentions
• You have available quota for the feature

On demand — from a text selection

You can also run relationship detection on a specific passage without saving:

1. Confirm at least 2 entity links in the text you want to analyse (unconfirmed suggestions don’t count)
2. Select the passage in the editor
3. Right-click → Extract Relationships, or press Ctrl+Shift+R
4. Suggestions appear in the Relationships tab immediately

This is useful when you want to catch relationships in one specific scene without triggering detection across your whole document, or when you want to run detection before you’re ready to save.

Note

The AI uses the relationship types defined in your world to constrain its suggestions. It only suggests relationship types that exist and are valid for the entity type pair involved. This significantly reduces false positives.

Data privacy

What’s sent: The text being analysed (up to 8,000 characters — either the full scene on save, or just the selected passage when triggered from a selection), the list of confirmed entities mentioned in that text, any existing relationships between those entities, and the catalogue of relationship types defined in your world (so the AI knows what types are valid).

What’s NOT sent: Your user identity, world ID, session IDs, or any data beyond what’s needed for relationship detection. Only the specific text being analysed is sent — not your entire content library.

Where it goes: To the AI model provider via OpenRouter, as configured in your model preferences.

Reviewing suggestions

Suggestions appear in the entity’s Relationships tab as pending items. Each suggestion shows:

• The suggested relationship type and direction
• The source and target entities
• A snippet of the text that triggered the suggestion (evidence text)
• The suggestion type (new or end)
• Confirm and Dismiss actions

You can confirm inline for straightforward suggestions, or open the full review modal for more control.

The review modal

Click a suggestion to open the review modal, which lets you adjust the AI’s interpretation before confirming:

Correcting entities — if the AI picked the wrong source or target, you can search for and select a different entity from your world without dismissing the suggestion.

Setting temporal bounds — you can set a valid_from and/or valid_to date on the relationship at confirmation time, using your world’s calendar system.

Changing the subtype — if the relationship type has subtypes (e.g. a family relationship might have subtypes like “parent”, “sibling”, “spouse”), you can select the correct one.

Adding extended properties — some relationship types have additional schema-defined fields. These appear as form inputs so you can fill them in while confirming.

Linking to a source scene — you can record which content unit this relationship was derived from, creating a provenance link between your lore and your writing.

Conflict resolution

If you’re confirming a “new” suggestion for a relationship type that has cardinality constraints (e.g. a character can only have one active allegiance at a time), the modal will show any conflicting active relationships of the same type.

For each conflict you choose:

• End — sets a valid_to date on the conflicting relationship (it ends when the new one begins)
• Delete — permanently removes the conflicting relationship
• Keep — confirms the new relationship without touching the existing one

You must make an explicit choice for each conflict before the confirm button becomes available.

Confirming a “new” suggestion creates the relationship in your knowledge graph. Confirming an “end” suggestion sets a valid_to timestamp on the existing relationship.

Accuracy

The AI is constrained in two ways:

1. Valid types only — it can only suggest relationship types that exist in your world and are valid for the entity type pair
2. Post-processing validation — invalid suggestions (wrong types, impossible pairings) are filtered out before they reach you

However, you should still review each suggestion — context and nuance matter, and the AI may occasionally misinterpret figurative language or ambiguous phrasing.