Free · Open source · Runs on your own machine
Point it at your D&D, Pathfinder or Call of Cthulhu notes and ask from Discord, Foundry's chat bar, or any MCP client. You bring the model — Vault does the wiring: finding the passages that bear on the question, holding the model to your notes, and citing the file every claim came from.
Which means you can ask it daft questions and still trust the answer.
Install it See what comes back
@Tusk who's the party member least fit to be the ship's captain?
Bryn Tallow. He cannot swim, has twice refused to board a boat — once at Dunmar, once at the Kelmoor quay — and gets sick on still watersource 1, which makes him the least seaworthy of the four regardless of any question of leadership.
Party.mdWhy this exists
My campaign notes were spread across six documents, two of which contradicted the other four. Players asked me the same questions between sessions — who was that merchant, what did we actually agree to, whose brother is the king — and I answered from memory, badly, and sometimes differently than last time.
The obvious fix is to ask an AI, and it fails in a specific way: a general chatbot handed your setting answers every question confidently, including the ones your documents say nothing about.
@Tusk what was the innkeeper at Greywater called?
A general chatbot
Her name was Mira Ashdown, and she had run the Drowned Lantern since her father passed.
No source. Nothing in your notes says this.
Tusk's Vault
I am unsure about this detail. I have recorded this as a lore gap for the DM to clarify.
Refused, and added to Lore Gaps
So the whole design is one idea: it may only tell you what your documents say, and it must show you where it got it. Answer a logged gap once and that answer is reused from then on.
Nobody needs this until the evening a player asks something specific and four people remember it four different ways. Then it is the only thing you want.
When it doesn't know
Any assistant can answer a question your documents cover. The one that decides whether you can trust it at the table is the question they don't — and the useful behaviour there is to stop, say so, and remember that you were asked.
Wednesday, in a channel
“Whose brother is the king?” — a question your notes never actually settled.
Two seconds later
Not a guess dressed as an answer. A fixed phrase, so nobody mistakes it for canon.
Same moment
The question lands in your dashboard, waiting for the one person who can settle it.
Once, ever
One sentence from you. That ruling is now part of the archive, cited like any document.
and every later phrasing of that question — however differently it is asked — gets your answer
A ruling recorded this way is cited like any other source, so six months later nobody has to remember whether it was decided or imagined. What people actually ask it →
Real questions, real citations, and one refusal, against a corpus written to be published. See the answers in full →
How it compares
Often you should. Most people arrive here having already pasted their notes into ChatGPT — good software, and for a question your documents genuinely answer it will usually answer correctly. The trouble is the question they don't.
NotebookLM is the closest comparison, and on its own ground it is better than this. Google's budget, excellent retrieval, exact citations, nothing to install. If you are working alone through a pile of documents, use it. The difference is not quality — it is what the two things are built for.
What a research notebook isn't trying to do
What this one charges you for that
| Capability | Tusk’s Vault | NotebookLM Google, in the cloud | ChatGPT / Claude a general assistant | Obsidian + AI plugins notes you already keep |
|---|---|---|---|---|
| At the table | ||||
| Everyone at the table sees the answer | ✓Yes | ✗No | ✗No | ✗No |
| Ask it from Discord, or Foundry’s chat bar | ✓Yes | ✗No | ✗No | ✗No |
| Answers in a voice your table picked | ✓Yes | ✗No | ◑Partly | ◑Partly |
| Whether it can play | ||||
| Will speculate when you let it, and labels it | ✓Yes | ✗No | ◑Partly | ◑Partly |
| Recounts a grim scene without sanding it down | ✓Yes | ✗No | ◑Partly | ◑Partly |
| Whether you can trust it | ||||
| Says it doesn’t know, instead of inventing | ✓Yes | ✓Yes | ✗No | ✗No |
| Cites the document behind every claim | ✓Yes | ✓Yes | ◑Partly | ◑Partly |
| A ruling you make once is remembered and reused | ✓Yes | ✗No | ✗No | ◑Partly |
| Your campaign documents never leave your machine | ✓Yes | ✗No | ✗No | ✓Yes |
| What it costs you | ||||
| Holds a campaign bigger than about 200 pages | ◑Partly | ✓Yes | ◑Partly | ◑Partly |
| Ready in five minutes, without a terminal | ✗No | ✓Yes | ✓Yes | ◑Partly |
| Still answers while your own machine is asleep | ✗No | ✓Yes | ✓Yes | ✗No |
| A team behind it, and somewhere to report a bug | ✗No | ✓Yes | ✓Yes | ◑Partly |
Which is the honest shape of it. This is not a better NotebookLM — it is a worse one, pointed somewhere NotebookLM is not pointed. Keep theirs for research; use this when the question is being shouted across a table. Campaign wikis and your own GM are not in that grid at all, because neither is a thing you ask — both, and the case for using them instead, are in the full comparison.
A grid where one column ticks every row is a grid whose rows were chosen by whoever owns that column. The four this one does not win are there so the nine it does are worth something. Read those four first.
How it works
A question goes in one end. Stage two is the one you cannot configure away, and the reason the answer at the far end can be trusted.
Reads a folder on your disk — PDF, DOCX, Markdown, plain text — or an Obsidian vault, read-only, nothing moved. Passage-level retrieval then picks out the parts that matter, so a large campaign is not a large bill.
Cite every claim; refuse rather than invent. The block is force-appended to every prompt, so editing the system prompt — or picking a persona — cannot remove it.
The reason this exists
Gemini, ~400 models through OpenRouter, a Claude subscription, or one running on your own machine. Swappable mid-campaign.
A refusal is labelled rather than served as content in the archivist's voice — and the question it could not answer lands in your dashboard as a lore gap.
The corpus travels with every question, so every cloud request pins a zero-data-retention routing floor. How the retrieval pipeline works →
Ask it anything
Most of what gets asked at a table is not "when did Dunmar fall". It is an argument two players are having, or a joke someone wants settled with evidence. Those work — and they work for the same reason the serious ones do. It can only answer from what you wrote down, and it shows you where.
@Tusk who's most likely to start a tavern fight?
Roswyn Kett is the likeliest — she has already put one person in the water after an argument with the harbourmaster at Kelmoor, and has since been barred from the Ninefold Rest twice (and readmitted twice)source 1.
Party.md@Tusk would Ser Alric get on with Vesper?
Unlikely, and pointedly so. Vesper has read every surviving account of the fall of Dunmar, and corrects people who have not — including, on one occasion, Ser Alric Vanesource 1. Ser Alric would not discuss Dunmar when the party found him at the Ninefold Restsource 2, and refuses to speak of the siege at allsource 3. A scholar who has already corrected the Grey Knight on the one subject he won't touch is not a natural friend for him.
Party.mdSessions.mdHouses.mdA persona replaces the whole voice, and the citation rules survive it. Same question, same corpus, same sources — two of the eight that ship:
@Tusk who's least fit to be the ship's captain?
The Archmage
Ah, a question of seamanship, not of secrecy — this one the chronicle answers plainly. Bryn Tallow is the least fit hand to stand at the helm. He cannot swim, and has twice refused to board a boat at all — once at Dunmar, once at the Kelmoor quay — and takes ill even on still water. A captain who will not set foot on his own ship is no captain at all.
Cited Party.md
The Sellsword
Bryn Tallow, and it ain't close — the poor bastard can't swim, gets sick on still water, and has flat-out refused to board a boat twice, once at Dunmar and once at the Kelmoor quay. Roswyn's the only one of the four who's ever actually crewed a boat, so hand her the wheel instead.
Cited Party.md
"Who would lose a duel to a chicken" is not in anybody's notes, so by default it is refused like any other question the documents don't answer. Turn on Speculative Mode and it will reason from what the chronicle establishes about each character — then cite the evidence it used and tag the conclusion, so a guess can never be mistaken later for something you wrote down.
@Tusk which of them is most likely to lose a duel to a chicken?
Speculative Mode off — the default
I am unsure about this detail. I have recorded this as a lore gap for the DM to clarify.
Refused. It isn't in your notes.
Speculative Mode on
Bryn Tallow — he's the one party member who's never drawn a weapon in any recorded session, can't swim, and has twice refused to even board a boat because still water makes him sick. A creature this jumpy and non-combative would fold against a chicken's raw aggression before a blade ever cleared its sheath.
Cited Party.md, tagged [speculation]
Being recorded
A video walkthrough is on the way
Install to first cited answer, start to finish — including the moment it refuses one.
Until it lands, the setup guide walks the same path in writing.
Getting a session in
A lore assistant nobody opens is a lore assistant nobody uses, so it answers where the question already gets asked.
Discord@mention the botCited answer
Between sessions, which is when most of the questions actually arrive. Mention the bot and the answer lands in the channel, citations and all, where everyone can see it — so the same question doesn't get asked three more times that week.
Wants a bot token. You paste one string; the rest is derived.
Foundry VTT/tusk your questionCited answer
Type /tusk and the answer posts into the chat log without anyone leaving the
game. A per-world access mode decides whether players may ask at all, and whether the
answer is whispered to them or posted for the table — because not every answer is
one the party should have.
Wants the companion module. Pairs in one click.
The third route needs no setup screen at all: any MCP client — Claude Code, Claude Desktop, anything speaking the protocol — can pair and ask directly. What people actually ask it →
And it doesn't have to be a campaign. Nothing in the machinery knows what a campaign is — it indexes a folder, cites what it finds, and refuses what isn't there. Point it at a handbook or a wiki export and you get the same bot.
The prompts, the citation styles and the refusal wording all assume a GM and a table. Everything else is a happy accident.
What it costs
Tusk's Vault is free. The only money involved is whatever your chosen model charges for the tokens, and a single question is a very small number of them — one retrieval-augmented request, not a generation pipeline. Gemini's free tier covers most home games outright, and a local model costs nothing at all.
On a Claude plan you already pay for
$0
On Gemini's free tier, which covers most home games
$0
On a model running on your own machine
$0
Through one OpenRouter key, a session of questions
pennies
A campaign wiki subscription, every month, whether you play or not
$5-15
Tusk's Vault
only when you played
A wiki subscription
every month, regardless
What leaves your machine
Years of somebody's worldbuilding, and often things said at a table that were never meant to leave it. So it's worth being exact about where everything goes. Almost all of it stays on the machine you installed it on.
Your machine
The model you chose
One person, one Windows PC, no support desk — which is why the feedback button sits next to the coffee one in the app. Telling me what broke gets it seen; it does not promise the fix sticks. The unvarnished version is here.
Worth asking of anything you hand years of worldbuilding. The answer splits cleanly.
Stays yours
No export step, because nothing was ever captured.
Stops
This is how it would actually die: drift, not deletion.
Questions
Start
You need Node 20 or newer and Git. That is the whole requirement — no Python, no Docker, no database, no admin rights, and no paid key: a free Gemini tier covers most home games, and a local model costs nothing at all.
Install it Connect it to Discord