Recursion Map Guide
A hands-on walkthrough of the recursion map: what the picture means, and how to use it to wire up (or tame) recursive lorebook entries. If you only want a quick reference, the Lorebook Editor page has a summary. This guide is the slow version.
How it works
What recursion actually is
In SillyTavern, a lorebook entry activates when the chat mentions one of its keys. With recursive scanning on, activated entries get a second life: the text of an activated entry is scanned too. So if lorebook entry A turns on, and A's content happens to contain one of entry B's keys, B can turn on as well. B's content then gets scanned, which can wake up C, and so on.
That's a chain reaction. Done well, it makes a book feel alive: mentioning a tavern pulls in the tavern keeper, whose entry pulls in her quest. Done badly, one common word snowballs into a wall of context you never asked for.
The recursion map draws those chain reactions for the current book so you can see them before they surprise you.
Reading the picture
Open the map from book settings (Map) or an entry's Options panel (Map). You get a full-screen view where:
- Each box is an entry.
- Each arrow is a possible unlock. An arrow from A to B means A's content mentions one of B's keys. Hover the arrow (or the boxes) for the actual key.
- Clusters are families. Boxes that can reach each other sit together. Separate clusters are independent: nothing in one cluster can wake up anything in another.
- Dots on a box are flags. The legend at the bottom decodes them: blue for non-recursable, green for prevent further, grey for delay until recursion, yellow for disabled entries.
Entries with no connections are hidden by default to keep the picture clean. Show standalone (top right) brings them in as dim boxes below the clusters. Useful for spotting entries you expected to be wired up but aren't.
Getting around
- Drag the background to pan, scroll to zoom toward your cursor.
- Reset view fits the whole book back on screen.
- Hover an entry to spotlight just its connections and fade the rest. Click to pin those paths and open the inspector. Click the same entry again to clear the pin so the whole map is even.
- The List button (top right) swaps the picture for a searchable table. Handy on very large books, which open in list mode automatically.
A first inspection
Click any entry. A panel slides in on the right showing three things:
- The entry's keys (as chips) and flags.
- Unlocked by: entries whose content mentions this entry's keys. Each row shows the matched keys as chips, with the line Mentions these keys in its content.
- Unlocks: entries this entry can wake up. Each row shows that entry's matched keys as chips, with the line These keys match in this content.
Click any entry in those two lists to inspect it instead. Walking those rows is the fastest way to follow a chain and check that each hop makes sense. Click the same box again, or the inspector X, to close the panel and drop the path highlight.
For quick fixes, edit the key chips or use the three flag buttons (Non-recursable, Prevent further recursion, Delay until recursion). Both write immediately.
Editing keys on the map
The chips are the entry's real primary keys. Changing them updates the lorebook entry and redraws the arrows. You do not have to close the map and go back to the Keys field, though that field stays in sync.
- Inspector Keys: add, rename, or remove any of this entry's primary keys.
- Unlocked by chips: same list as inspector Keys (this entry owns them).
- Unlocks chips: the other entry's keys. Remove or rename one to break or retarget that hop without opening that entry first.
Click a chip to rename it. Enter or blur saves; Escape cancels and leaves the map open. X removes it. Type next to the inspector chips and press Enter or comma to add a key.
These are the same keys SillyTavern uses to activate the entry from chat, not aliases for the picture. Secondary (selective) keys are not on the map.
The camera stays where you left it when arrows appear or vanish. Reset view recenters if a new cluster lands off-screen.
Flags
The three flags, in plain terms
| Flag | What it does in the chain |
|---|---|
| Non-recursable | This entry is a dead end for incoming chains. Other entries can name its keys all they want; it will not turn on through recursion. It can still activate normally from the chat. |
| Prevent further recursion | When this entry activates, it unlocks nothing. A chain stops here even if its content mentions other keys. |
| Delay until recursion | This entry ignores the first scan of the chat and only turns on via recursion. Good for "deep lore" that should only surface once a topic is already rolling. |
Two combinations cover most needs:
- Standalone entry: non-recursable + prevent further. Fully out of the web; activates only from chat and chains nowhere.
- Seed entry: prevent further off, non-recursable on. Starts chains but nothing can recursively start it. Typical for hub entries like a location.
Changing many entries at once
Sometimes a whole cluster needs taming. That's what selection is for:
- Select entries. Ctrl (or Cmd) + click boxes on the web, or use the checkboxes in list view. Or use the shortcuts in the selection bar: All, Visible (current search results), or Linked (everything that has at least one path).
- Stage a change. In the bar that appears at the bottom, press On or Off next to a flag, or press Isolate: block both directions. Nothing happens yet. The bar shows exactly what is staged, which direction, and for how many entries.
- Commit or back out. Press Apply to write the change, or Discard to drop it. Staging something else, or changing the selection, clears the pending change. On selections above 25 entries, Apply asks for a second confirmation click.
The map does not rearrange itself while you work. Flag changes don't add or remove paths (paths come from keys and content), so boxes stay put; only the flag dots update.
Common repairs
A chain explodes into everything
Inspect the entry where the explosion starts and look at its "Unlocks" list. Usually one broad key (a name, a place) is matching half the book. Tighten that target's keys on the hop chips, or set Prevent further recursion on the entry that fans out.
Two entries keep re-triggering each other
A mentions B's key, B mentions A's key. Flags can't break the loop by themselves. If the hop is a key you do not actually want, remove that chip on the Unlocks row. If the word should stay as a trigger but should not appear in the other entry's content, close the map and edit that prose.
An entry never fires recursively
Click it and check "Unlocked by". Empty? Either nothing mentions its keys (rename the keys in the inspector chips, or mention them where you want), or Non-recursable is on, which the panel will say directly.
Deep lore fires too early
Set Delay until recursion so it stays quiet on the first chat scan and only joins once the topic is already active.
What the map does not know
The picture is built from primary keys in content. It deliberately ignores:
- secondary (selective) keys and their logic
- activation probability
- token budgets that might cut an entry mid-chain
- how many scan steps SillyTavern actually runs
So the map shows what can connect, not what will fire in a specific chat. It's for shaping the wiring, not predicting a single conversation. If Recursive scanning is off for the book, the map shows potential connections and reminds you that SillyTavern won't follow them until scanning is enabled.
Next steps
- Lorebook Editor - fields, entry editor, reference for the map
- Lorebook Vault - managing standalone books
- Snapshots & Rollback - snapshot a book before big flag sweeps