Entity Detection & Mentions

As you write in the content editor, MythTapestry can detect references to entities in your text and suggest linking them to your knowledge graph. This feature is entirely optional — you can always link entities manually using the @ mention system instead.

How it works

1. You write content that mentions an entity (e.g. “Aldric rode to the Crimson Tower”)
2. The system detects “Aldric” and “Crimson Tower” as potential entity references
3. Suggestions appear as highlighted text with a subtle indicator
4. You can confirm or dismiss each suggestion

In the scene editor, detection lives on the Entities tab of the right-hand panel. Click Detect Entities to run it.

Detection methods

There are two ways to run detection — you choose per world in World Settings.

Local (no AI, no quota)

The local detector runs entirely on the MythTapestry server. Nothing about your writing leaves the server, no AI provider is involved, and it doesn’t draw from any quota.

• Fast — usually under a second per scene
• Private — text stays on the server; no third-party AI sees it
• Free — doesn’t use any allowance
• Good for straightforward references: proper names, places, organisations you’ve already added to your world
• Less good at unusual phrasings, common-word names (“The Crown” as a faction), or anything that needs to read context

Data privacy — local mode

What’s processed: the text in your scene, plus the names, types, and aliases of entities you’ve added to this world.

Where it runs: on the MythTapestry server only. No external AI provider is involved.

Planned: a future “in-browser” mode that would run the detection inside your browser, so the text never leaves your device at all.

AI

The AI detector sends the text to a language model (the one you’ve picked in model preferences) along with the list of entities in your world. It’s better for context-dependent references and ambiguous phrasings.

• More accurate for unusual or ambiguous references
• Better at spotting that “The Crown” refers to a faction, not a literal crown
• Uses your Entity Detection quota

A run does more than name things. It also checks whether a suggestion is somebody you already have under a slightly different spelling — so a second Kaelen Thorne doesn’t quietly appear beside the first — and, where a mention was matched to the wrong entity, proposes the one it thinks you meant. Those checks are part of the same run and don’t cost anything extra: one detection, one operation, however much thinking happens inside it.

Data privacy — AI mode

What’s sent: the text being analysed, plus the names, types, and aliases of entities in this world (so the model knows what it’s allowed to flag).

What’s NOT sent: your user identity, session IDs, world ID, images, files, relationship data, or any data from your other worlds.

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

What we record afterwards: a count of how many suggestions the model returned and how many were discarded for breaking your world’s rules, together with which model produced them. Counts only — no names, no text, and nothing tied to you or your account. It tells us which models to trust with your writing.

Detection modes

Mode	Trigger	Description
Manual	Click the “Detect Entities” button	You control when detection runs
Auto	On content save	Detection runs automatically when you save
Selection	Ctrl+Shift+E or right-click → Extract Entities	Runs on the selected text only

The manual and auto modes are configured per world in World Settings. You can disable detection entirely if you prefer to manage mentions manually.

Selection-based extraction

Instead of scanning your whole document, you can extract entities from a specific passage:

1. Select the text you want to analyse in the editor
2. Press Ctrl+Shift+E, or right-click → Extract Entities
3. Only the selected text is sent through the detection pipeline
4. Results appear in the Entities tab as suggestions, ready to confirm or dismiss

This uses exactly the same GLiNER or AI pipeline as full-document detection — just scoped to your selection. It’s particularly useful for dense passages (introductions, infodumps, battle scenes) where you want to catch every reference without re-scanning work you’ve already reviewed.

Note

Detection must be enabled in World Settings for selection-based extraction to work. The keyboard shortcut is silent if detection is disabled.

Managing mentions

Single mention

When one entity is detected in a text selection, you see an inline confirm/dismiss action right at the mention location.

Multiple mentions

When several entities are detected, the Entities tab sorts them into two groups, and each group’s buttons act on that group only:

• Smart Suggested — mentions of entities you already have. Confirm All links every one of them; Reject All removes them. The badge counts the entities in this group.
• New Entities — things that look like entities your world doesn’t have yet. Create All creates each one and links its mentions; Dismiss All removes them without creating anything.

Confirm All, Reject All and Dismiss All ask first, and say how many suggestions they will touch. Rejecting or dismissing can’t be undone. You can still confirm or dismiss each mention on its own.

Where a confirmed link goes

Each suggestion remembers the exact words it was found in, and confirming it links those words, not the first match on the page. Say “Alaric” is suggested in your second paragraph while the first already reads @Alaric Dragonheart: confirming links the second paragraph’s “Alaric” and leaves the existing link whole. Detection also skips any words already inside a link.

Pronouns are never offered as links. Detection reads “she” and “him” to work out who is who, so She waved him through still counts towards Merrow Vane, but it won’t suggest turning “She” into a link.

Creating a suggested entity checks your world first. If the AI offers “Osric Vane” as new and you already have an Osric Vane, the mention links to yours rather than making a second one. This applies to the quick create button; Review & edit creates exactly what you fill in.

Tip

You can also manually mention an entity by typing @ in the editor and searching for it. Manual mentions are always confirmed immediately — no detection system needed. Those you type into a new piece before its first save count from the moment you click Create.

Mention states

State	What it looks like	What it does
Confirmed	Solid highlight, clickable link to the entity	Stays across saves; shows up in the graph and on the entity’s Mentions tab
Suggested	Subtle highlight with confirm/dismiss actions	Stays across saves but does not appear in the graph or count toward features that need confirmed mentions
Dismissed	Removed from the text	Won’t be suggested again at this spot

Note

Suggestions are preserved when you save — you don’t have to deal with them in one sitting. They simply don’t “count” toward features that look for confirmed mentions (like Relationship Suggestions or Entity Assist’s Suggest Updates) until you confirm them.

Suggesting new entities

If you enable Suggest new entities in World Settings, detection can also flag text that looks like it refers to an entity that doesn’t exist in your world yet. These appear with a different indicator so you can decide whether to create the entity then and there. This pairs well with Entity Assist’s Fill with AI, which can pre-fill the new entity’s fields from the same scene in one step.

Detection sensitivity

The confidence threshold (default 0.7) controls how sure the detector has to be before it flags something. Lower it to catch more (at the cost of more false positives), raise it to be stricter (at the cost of missing some). Most worlds work fine on the default; only adjust if you’re seeing too many or too few suggestions in practice.