How LoreMonkey uses AI

Last updated 24 September 2026

The two kinds of AI in it

• Speech recognition, on your PC. A Whisper speech model runs on your own computer and turns what is said at the table into a transcript. It writes down what it hears rather than inventing anything, but it mishears, especially names, so a transcript is a close record rather than an exact one.
• A language model, over the internet. Claude models from Anthropic, reached through LoreMonkey's server, read extracts of your transcript and notes and write text back. The privacy policy says exactly what is sent and what is kept.

What the language model writes

• Suggestions during play: the prompts in the Suggestions panel, and the desktop notifications that announce them.
• Ask LoreMonkey: every answer. When you ask a question you are talking to an AI, not to a person, and nobody at LoreMonkey reads the conversation.
• Session recaps.
• Campaign memory: the summaries of past sessions that Ask and the suggestions draw on.
• Drafted changes to your notes: updates after a session, campaign files written from your documents, rebuilt files, and new compendium entries drafted from what your notes already say.
• Finding names: characters and places your notes missed, picked out of your transcripts.

How you can tell in the app

We kept this quiet on purpose: one small label where AI text appears, rather than a warning on every line.

• The Suggestions panel, the session recap and Ask LoreMonkey carry an AI label beside their heading, and Ask says in so many words that an AI is answering.
• Changes waiting for your review say Drafted by AI or Rewritten by AI, and so does a drafted compendium entry.
• A desktop notification for a suggestion is titled LoreMonkey AI suggestion, because you read it away from the panel.

A mark software can read

When you export a campaign to an Obsidian vault, every session recap carries loremonkey-ai-generated: true in its frontmatter, so a tool reading the vault can tell it began as an AI's writing. Importing the vault back takes the mark off again, along with the rest of what LoreMonkey added. Your own campaign notes carry no such mark, because they are yours even where you accepted an AI's suggestion into them.

Text you copy out of the app is plain text and cannot carry a hidden mark. If you post a recap somewhere other people will read it, saying it was written with AI is the honest thing to do, and we would encourage it.

You stay in charge of your notes

An AI's changes to your campaign files are proposals. They wait on the Character & Campaign tab until you apply them, and nothing is written until you do. Campaign memory is the one thing the AI writes directly, so that Ask can use a session straight away; every memory record can be corrected, switched off or deleted on the same tab, and a correction never changes the transcript it came from.

It can be wrong

A language model can misread a transcript, miss something, or state a detail your notes never said. Ask LoreMonkey answers only from your campaign's own records and says when they do not cover a question, but treat everything the AI writes as a prompt for your own judgement. It is a game aid, as clause 8 of the terms says.

What we do not use AI for

• We do not train models on your sessions or notes, and our model provider does not use them for training by default.
• LoreMonkey does not recognise anyone by their voice. On Discord, who is speaking comes from the Discord client already on your PC; at an in-person table LoreMonkey does not label speakers at all.
• It does not generate images, audio or video, and it does not imitate anyone's voice.
• No AI decides anything about your account, your plan or your bill. Those follow fixed rules.

Questions

Write to hello@loremonkey.com.Scott Jacobs, trading as LoreMonkey, is the provider of LoreMonkey.