Databank¶

The Databank is Lumiverse's document knowledge base. Drop reference material into a databank — worldbuilding notes, character backstories, rulebooks, transcripts, scraped articles — and it becomes retrievable context during chat. Documents are chunked, vectorized, and either pulled in automatically based on what's being said, or summoned on demand with a #slug mention.

Think of it as a per-user research folder that the AI can flip through while it answers.

How It Differs From World Books and Long-Term Memory¶

World Books are for short, structured lore facts. Databanks are for long-form source material you want the AI to consult — chapter PDFs, design docs, full character bibles, transcripts.

Scopes¶

Every databank belongs to one of three scopes:

A chat sees all currently-attached databanks at once — global + character-bound + chat-bound — so you can layer them.

Attaching a Databank¶

1. Open the Databank drawer tab
2. Use the Global / Character / Chat scope tabs at the top of the panel
3. Pick the databank you want from the list, or click Attach to bind an existing global databank into the active character or chat

Chat documents auto-create a chat-scoped databank the first time you upload a file in that chat — no manual setup required.

Uploading Documents¶

From the Databank panel, with a bank selected:

1. Click Upload (or drag files onto the panel)
2. Files are parsed, chunked, and embedded in the background

Each document shows a status badge while it's being processed:

Supported Formats¶

.txt, .md, .markdown, .csv, .tsv, .json, .xml, .html, .htm, .yaml, .yml, .log, .rst, .rtf

PDF, DOCX, EPUB, audio, and image OCR are not supported — convert those to Markdown or plain text first.

File Size¶

Each file is capped at 10 MB. There's no overall storage cap, but very large databanks (thousands of long documents) will use more disk and slow embedding rebuilds.

Scrape a URL¶

Instead of downloading-then-uploading, paste a URL into the Paste a URL to scrape… field at the top of the document list. Lumiverse fetches the page, extracts the readable content, and ingests it as a new document.

Slug Mentions (#document-name)¶

Every document gets an auto-generated, kebab-cased slug based on its name:

• My Worldbuilding Notes.md → #my-worldbuilding-notes
• Chapter 12 - The Reckoning.txt → #chapter-12-the-reckoning

Type # in the chat input and an autocomplete popover lets you pick a document from any attached databank. The mention is removed from the message before it's sent — the AI sees the document content instead.

Sizing behavior: if a full document fits within a 2,000-token budget, the entire text is injected. If it doesn't, Lumiverse runs a semantic search inside the mentioned document and injects the most relevant chunks. Either way, the inserted content is labeled with [Source: <document name>] so the AI knows where it came from.

Renaming changes the slug. If you rename a document, update any presets or chat templates that reference it.

Automatic Retrieval¶

When you generate a message, Lumiverse also runs a semantic search across every attached databank using your recent chat context as the query. The top matching chunks are pulled in as additional context.

This requires:

• Embeddings configured (any supported provider)
• At least one attached databank with ready documents

If embeddings aren't set up, documents are still parsed and chunked — but only #slug mentions work, since there's no way to run semantic search without a vector model.

Settings¶

The Databank panel exposes a small settings section that applies globally to every databank you own:

After changing chunk parameters, hit Reprocess All on any existing documents — old chunks aren't automatically resized.

Document Operations¶

For each document, the panel offers:

• Rename — click the title to edit inline (updates the slug too)
• Edit content — click the body to open it in the inline editor; saved edits trigger re-chunking
• Reprocess — re-parse, re-chunk, and re-embed a single document
• Delete — remove the document and its chunks

The whole bank has:

• Reprocess All — useful after changing chunk settings
• Fuse — merge one databank into another. The source's documents are moved over, duplicates are collapsed by content hash, and any characters/chats attached to the source are rewired to the target. Handy for consolidating "my fantasy world" notes that ended up split across several banks.

Macros¶

Databank content is available in your presets through dedicated macros:

Aliases for {{databank}}: {{databankMemory}}, {{documents}}, {{knowledgeBank}}.

Use {{databankActive}} to wrap databank content in a conditional so the section disappears cleanly when nothing retrieves:

{{if::{{databankActive}}}}
Relevant source material:
{{databankRaw}}
{{/if}}

Tips¶