A VtM relationship map is easy to draw and hard to keep current

Keeping a VtM relationship map current by hand

What goes wrong with a hand-drawn relationship map is the maintenance, not the drawing. A few habits keep a paper or whiteboard map honest for longer, whatever tool you end up using:

• Date every line with the session it was drawn in, so a stale tie looks stale.
• Strike through a relationship that turned instead of erasing it. In a Vampire chronicle the history of a tie is half of what it means.
• Use arrows, not lines, for anything that runs one way — a debt, a loyalty, a grudge the other party does not know about.
• Keep two layers, what the coterie knows and what only the Storyteller knows, on separate sheets or in separate colours.
• Update it straight after the session, in five minutes, while the table is still arguing about what just happened — not before the next one, when it has gone soft.

A V5 relationship map built from what the coterie actually says and does

That last habit is the one that fails, because it is homework at the end of a long night. Renegade Game Studios sells Vampire as "a Storytelling game of personal and political horror" (see the table below), and here the political half is bookkeeping done for you: after each session, the chain that builds the wiki and the recap reads the transcript for connections between the people, places and factions in your wiki and adds them to a graph (how the relationship map works).

It is the tool this page is about, and it is not free. The map needs the Dungeon Master plan ($7.99 a month) or above; on the free plan, detection does not run at all. A free account still gets the transcript, the wiki and the recap — the material the map is drawn from — and if you upgrade later, the map can be generated from the sessions you already transcribed.

To seed it, record the session where your table builds its web out loud, or type a paper map up as plain sentences — who sired whom, who owes whom — and bring it in with transcript import, which spends no transcription hours (two imports a month free, ten on Dungeon Master, unlimited on Guild Master). There is no way to upload a photo of a chart.

The map stays yours. The GM notes written for detected connections — the prompt behind them asks for secrets and betrayals — are stripped from anything a player in your campaign can read, and a shared wiki link (also Dungeon Master and up) gives players the wiki, not your map.

Sires, boons and blood bonds are not on the label list

When the map detects a connection, the model is asked to label it with a generic, D&D-shaped category: family, allied, enemy, works for, member of, owns, knows, located in and the like. None of them says sire, owes a boon to or blood-bound to, so the ties a Vampire chronicle turns on are likely to arrive under the nearest generic word or not at all. Detection also writes one connection per pair: a sire who is also a rival is one line, not two.

The fix is the GM editor. Add the connections the model missed and correct the ones it got wrong: choose from eleven types — Parent Of and Child Of are the nearest thing to sire and childe, Serves fits a ghoul — set the strength, mark a new connection one-way or mutual (a boon owed rarely runs both ways), and put the Vampire meaning in the description, up to 500 characters. Connections you add, and detected ones you confirm, are left alone by every later session’s analysis.

What each session changes, and what it leaves to you

Each session adds what is new, and for connections already on the map it refreshes the evidence — how often, and in how many sessions, the pair has come up — rather than freezing it at the session that first found them.

What it does not do is rewrite a connection when the story turns it. A pair first read as allies keeps that label and description after the betrayal until you change it; re-authoring every connection every session would multiply the most expensive step in the pipeline by the size of your graph. The map keeps up with who is connected; you keep up with what each connection now means. After a night when the Primogen turned on each other, that is a few minutes in the editor.

One limit matters more in a game this full of in-character dialogue: the transcript does not say who spoke. "I will never forgive you, Marcus" carries Marcus’s name but not the speaker’s, so the model has to infer who "I" is from context and may not manage it. Ties the Storyteller narrates, or the table states in the third person, give it both names.

Nothing in the pipeline is rules-aware, and the prompts say D&D

Transcription turns speech into text; extraction pulls out the people, places, items and factions that were mentioned; the recap summarises what was said; the map connects them. No step reads a rulebook, tracks Hunger or Blood Potency, or knows what a Discipline does — and there are no character sheets, dice pools or city maps here.

The AI is also told, in plain text and at more than one stage, that it is looking at a D&D game. The transcription prompt opens "A Dungeons and Dragons tabletop session"; extraction asks for entities from a "D&D session transcription"; and the prompts behind the relationship map and the recap call what they read a D&D session too. We would rather show you that wording than have you find it later. For Vampire the cost is mostly vocabulary: street names, surnames and nightclubs plus the game’s own terms, none of which a D&D framing helps with.

One more place the fantasy assumption shows is the off-topic detector, which looks for real-world table talk that slipped into the wiki. Its free check runs whenever you open the campaign, and it is a fixed keyword list that knows nothing about your setting, so a chronicle set in a real city will see entries like Chicago, or a character whose description mentions the internet or a cell phone, marked as possibly off-topic. The marks are suggestions and nothing is removed unless you remove it: press Keep on anything that belongs and it will not be flagged again. Before removing anything from that screen, untick "Also block these names from future AI extraction", or you can put your own city on the AI ignore list. The optional AI deep scan takes a sample of your own wiki entries as its sense of the setting, but it is also told to expect high fantasy.

Clan, sect and domain names belong in the glossary

Each campaign has a glossary you control, fed into transcription ahead of every session alongside your player characters and recent wiki names. Put the chronicle’s proper nouns in it — the coterie, the Prince and the city’s other Kindred, clan and sect names the way your table says them, the club that serves as Elysium — and they come back spelled consistently instead of phonetically. Here that matters twice: a name misspelled in the transcript can become a second wiki entry, and a second dot on the map, with its connections split between the two.

Entries take pronunciation hints too, rendered like "Wendelbrie (WEN-del-bree)". There is a budget — the transcription API honours about the last 224 tokens of a prompt, so the glossary is budgeted in tokens (a few hundred characters of names in practice, 850 at the absolute most) — so it is a list of your most-confused names, not every Kindred in the city. The Whisper guide has the underlying detail.