DeepLore - Obsidian Vault Lorebook for SillyTavern

Surface the right lore from your vault at the right moment. DeepLore connects your Obsidian vault to SillyTavern, automatically injecting relevant world-building notes into AI prompts when keywords appear in conversation.

Upgrading? Make sure to install the new server index.js. New since 0.10: shared core refactor (no behavior changes), active character boost. See the changelog for details.

Features

• Keyword-triggered injection -- Tag Obsidian notes with keywords via YAML frontmatter. When keywords appear in chat, the note content is injected into the AI prompt automatically.
• Always-send and never-insert tags -- Force critical lore to always be present or mark draft notes to be skipped.
• Recursive scanning -- Matched entries are scanned for keywords that might trigger additional entries, building chains of related lore.
• Token budget controls -- Set limits on how many entries or tokens get injected per generation.
• Configurable injection position -- Inject before/after the system prompt, or in-chat at a specific depth as any role.
• Vault review command -- Send your entire lorebook to the AI for consistency review with /deeplore-review.
• Per-entry overrides -- Set custom scan depth, priority, and recursion behavior per note via frontmatter.
• Per-entry injection position -- Override the global injection position, depth, and role on a per-entry basis via frontmatter.
• Conditional gating -- Entries can declare dependencies (requires) and blockers (excludes) on other entries.
• Cooldown & warmup tags -- Per-entry cooldown skips injection for N generations after triggering. Per-entry warmup requires N keyword occurrences before first trigger.
• Re-injection cooldown -- Global setting to skip re-injecting entries for N generations after last injection, saving context.
• Vault change detection -- Detects added, removed, and modified entries when the index rebuilds, with optional toast notifications.
• Active character boost -- Optionally auto-match the active character's vault entry by name, ensuring their lore is always available when they're in the conversation.
• Entry analytics -- Track how often each entry is matched and injected. View with /deeplore-analytics.
• Entry health check -- Audit entries for common issues (empty keys, orphaned requires/excludes, oversized, duplicate keywords) with /deeplore-health.
• World Info interop -- Optionally let SillyTavern's built-in World Info scan injected lore for cross-system triggering.

Prerequisites

• SillyTavern (1.12.0+)
• Obsidian with the Local REST API community plugin installed and enabled
• Server plugins enabled in SillyTavern (enableServerPlugins: true in config.yaml)

Installation

Step 1: Install the client extension

Use SillyTavern's built-in extension installer (recommended):

1. Open SillyTavern
2. Go to Extensions panel > Install Extension
3. Paste this URL: https://github.com/pixelnull/sillytavern-DeepLore
4. Click Install

Or install manually with git:

cd SillyTavern/data/default-user/extensions
git clone https://github.com/pixelnull/sillytavern-DeepLore.git

Step 2: Install the server plugin

DeepLore needs a server plugin to talk to Obsidian. Re-run this step after every update -- the server plugin is not updated automatically when you pull new extension code.

Option A: Use the installer script (recommended)

Run the installer from the extension directory:

• Windows: Double-click install-server.bat or run it from the command line
• Linux/Mac: Run ./install-server.sh

If the extension isn't installed inside SillyTavern's directory, pass the SillyTavern root path as an argument:

./install-server.sh /path/to/SillyTavern

Option B: Manual copy

1. Find the server folder at SillyTavern/public/scripts/extensions/third-party/sillytavern-DeepLore/server
2. Copy it into SillyTavern/plugins/
3. Rename it to deeplore

The result should be: SillyTavern/plugins/deeplore/index.js

Step 3: Enable server plugins

In your SillyTavern config.yaml, set:

enableServerPlugins: true

Step 4: Restart SillyTavern

Restart the SillyTavern server so it picks up the new plugin, then refresh the browser.

Setup

1. In Obsidian, install and enable the Local REST API community plugin
2. Note the API port (default: 27123) and copy the API key from Obsidian Settings > Local REST API
3. In SillyTavern, go to Extensions > DeepLore
4. Enter the port and API key, then click Test Connection
5. Check Enable DeepLore
6. Click Refresh Index to pull your vault entries

Writing Lorebook Notes

Tag any Obsidian note with #lorebook (configurable) and add a keys field in the YAML frontmatter:

---
tags:
  - lorebook
keys:
  - Eris
  - goddess of discord
priority: 10
---

# Eris

Eris is the goddess of discord and strife. She carries a golden apple
inscribed "To the Fairest" which she uses to sow chaos among mortals
and gods alike.

Frontmatter Fields

Field	Type	Default	Description
tags	array	(required)	Must include your lorebook tag (default: lorebook)
keys	array	[]	Keywords that trigger this entry when found in chat
priority	number	100	Sort order (lower = injected first)
constant	boolean	false	Always inject regardless of keywords
enabled	boolean	true	Set to false to skip this note
scanDepth	number	(global)	Override the global scan depth for this entry
excludeRecursion	boolean	false	Don't scan this entry's content during recursive matching
requires	array	[]	Entry titles that must ALL be matched for this entry to activate
excludes	array	[]	Entry titles that, if ANY are matched, block this entry
position	string	(global)	Injection position override: before, after, or in_chat
depth	number	(global)	Injection depth override (for in_chat position)
role	string	(global)	Message role override: system, user, or assistant
cooldown	number	(none)	After triggering, skip this entry for N generations
warmup	number	(none)	Require keyword to appear N times before triggering (must be >1)

Special Tags

• #lorebook -- Marks a note as a lorebook entry (configurable in settings)
• #lorebook-always -- Forces the note to always be injected, like constant: true
• #lorebook-never -- Prevents the note from ever being injected, even if keywords match

Slash Commands

Command	Description
/deeplore-refresh	Force rebuild the vault index cache
/deeplore-status	Show connection info, entry counts, and cache status
/deeplore-review [question]	Send all entries to the AI for review. Optionally provide a custom question.
/deeplore-analytics	Show entry usage analytics: match and injection counts per entry
/deeplore-health	Audit entries for common issues (empty keys, orphaned references, oversized, duplicates)

Settings Reference

Connection

• Obsidian API Port -- Port for the Local REST API plugin (default: 27123)
• API Key -- Bearer token from Obsidian's Local REST API settings

Vault Settings

• Lorebook Tag -- Tag that identifies lorebook notes (default: lorebook)
• Always-Send Tag -- Tag for entries that always inject (default: lorebook-always)
• Never-Insert Tag -- Tag for entries that never inject (default: lorebook-never)
• Scan Depth -- How many recent messages to scan for keywords (default: 4)
• Max Entries / Unlimited -- Cap on injected entries per generation
• Token Budget / Unlimited -- Cap on total injected tokens per generation

Matching

• Case Sensitive -- Whether keyword matching respects case
• Match Whole Words -- Use word boundaries so "war" won't match "warning"
• Active Character Boost -- Auto-match the active character's vault entry by name or keyword
• Recursive Scanning -- Scan matched entry content for more keyword triggers
• Max Recursion Steps -- Limit on recursive scan passes (default: 3)
• Re-injection Cooldown -- Skip re-injecting an entry for N generations after it was last injected (0 = disabled)

Injection

• Injection Template -- Format string with {{title}} and {{content}} macros
• Injection Position -- Where in the prompt to insert lore (before/after system prompt, or in-chat at depth)
• Allow World Info Scan -- Let ST's World Info system scan injected lore

Index & Debug

• Cache TTL -- How long (seconds) to cache the vault index before re-fetching (default: 300)
• Review Response Tokens -- Token limit for /deeplore-review responses (0 = auto)
• Auto-Sync Interval -- Seconds between automatic vault re-checks (0 = disabled). Detects changes without manual refresh.
• Show Sync Change Toasts -- Show toast notifications when vault changes are detected
• Debug Mode -- Log match details to browser console (F12)

How It Works

1. On each AI generation, the extension scans the last N chat messages for keywords
2. Notes whose keys match are collected, sorted by priority, and trimmed to budget
3. Matched content is formatted with the injection template and inserted into the prompt
4. If recursive scanning is on, matched entries are scanned for keywords that trigger more entries
5. The vault index is cached and refreshed automatically based on the Cache TTL

License

MIT