-
Notifications
You must be signed in to change notification settings - Fork 8
Home
Welcome to DeepLore. It keeps your stories grounded in your worldbuilding by feeding the right lore to the AI exactly when a scene needs it — drawn straight from your Obsidian vault.
If you've used SillyTavern's World Info (lorebooks), you know the pain: keyword matching breaks down as your world grows, and the entry you needed stays cold because nobody typed the magic word. DeepLore fixes that. It runs keyword matching first, then asks an AI to read what survived and pick the entries that actually fit the scene — so the right lore fires even when the keyword was never typed.
It plugs into SillyTavern and works in the background while you write. As you roleplay, the writing AI can flag gaps in your world mid-generation, and Emma — the Librarian agent — helps you fill them in. Your story builds your world; your world flows back into your story.
New here? Start with What is DeepLore, then Quick Start to get it running in five minutes.

Note
The older standalone sillytavern-DeepLore extension is deprecated. This repo (formerly DeepLore) is the active project. Do not run both at once. Running both corrupts prompt injection.
- Three pipeline modes: two-stage (keywords pre-filter, then AI selects), AI-only (full vault sent to AI, no keyword filtering), and keywords-only (no AI). Two-stage is the default.
- Any LLM provider: Anthropic, OpenAI, OpenRouter, Gemini, DeepSeek, local backends (Ooba, KoboldCpp, llama.cpp), and anything else SillyTavern's Connection Manager supports.
- Multi-vault support: connect multiple Obsidian vaults at once with content-hash dedup.
-
The Librarian (Emma): writing-AI tools (
search,flag) that detect missing lore mid-generation, plus Emma, a chat agent who helps you author the entry. Seelorebook-guidefor Librarian-only writing guides. - Per-chat pin and block: override entry selection per chat. Force-include or exclude specific entries.
-
Contextual gating: filter entries by era, location, scene type, characters present, and any custom field you define in
field-definitions.yaml. Visual rule builder included. - Context Cartographer: per-message trace showing which entries fired and why (keyword hit, AI selection, constant, cascade, pin).
- Session Scribe: auto-summarize sessions back to your Obsidian vault as timestamped notes.
- Auto Lorebook: AI suggests new entries from chat content; accept or reject each one.
- Author's Notebook and AI Notepad: per-chat scratchpads (one human-written, one AI-written) injected as context.
-
World Info import:
/dle-importconverts SillyTavern World Info JSON exports into vault entries. - Relationship graph: force-directed layout, 200+ nodes, Louvain clustering, gap analysis, focus mode.
- Fine-grained matching: cooldown, warmup, probability, refine keys, cascade links, per-entry injection overrides, entry decay.
- Bootstrap entries: seed and bootstrap injection for early conversations on new chats.
- Persistent infrastructure: IndexedDB cache, reuse-sync indexing (skips re-parse of unchanged entries), AI search circuit breaker, prompt cache optimization.
- Diagnostics: analytics, health check, pipeline inspector, activation simulation, "Why Not?" tracing, entry browser.
- SillyTavern 1.12.14+
- Obsidian with the Local REST API plugin enabled
- A lore vault. Your existing Obsidian vault works;
/dle-importconverts World Info JSON if you're migrating. - For AI search: any LLM provider via a SillyTavern Connection Manager profile. Keywords-only mode runs without any provider.
| Page | Description |
|---|---|
| What is DeepLore | What a lorebook is, why DeepLore, and how the pipeline works (start here) |
| Installation | Step-by-step setup guide |
| Quick Start | 5-minute getting started guide |
| First Steps | Building your vault and tuning the extension |
| For World Info Users | Field-by-field cheat sheet for migrating from World Info |
| Features | Linked catalog of all features |
| Feature pages | |
| Drawer | Live side panel: Why?, Browse, Gating, Librarian, Tools tabs, temperature heatmap |
| Inspection and Diagnostics | Context Cartographer, "Why Not?" tracing, health check, simulation |
| AI-Powered Tools | Session Scribe, Auto Lorebook, Optimize Keys, Auto-Summary |
| Entry Matching and Behavior | Cooldown, warmup, gating, fuzzy search, decay, cascade links |
| Injection and Context Control | Injection positions, Prompt Manager integration, pin and block, contextual gating |
| Custom Fields | User-defined frontmatter fields for contextual gating |
| AI Notepad | AI-written private session notes; tag mode and extract mode |
| Infrastructure | Multi-vault, IndexedDB cache, circuit breaker, AI caching |
| Setup and Import | Setup wizard, quick actions, World Info import |
| Setup Wizard | First-run configuration wizard |
| Reference | |
| Writing Vault Entries | How to author lorebook entries with copy-paste templates |
| AI Search | How AI selection works inside the pipeline |
| Pipeline | The full retrieval and gating flow |
| Settings Reference | Every setting documented |
| Slash Commands | All slash commands |
| Glossary | Key terms and definitions |
| Troubleshooting | Common issues and fixes |
| FAQ | Frequently asked questions |
DeepLore Wiki
Getting started
Features
Reference
Help