Settings Reference

Settings reference

Every DeepLore setting, with default, range, and effect. Values match settings.js defaults; the constraints column matches settingsConstraints (numeric clamp ranges enforced on every load).

Important

v2.5: Custom Proxy mode removed

Earlier versions let any AI feature route through a "Custom Proxy" connection mode (ST's CORS proxy to an external endpoint such as claude-code-proxy). That mode was removed in v2.5. Every DLE AI feature now uses Connection profile (or, where supported, inherits another feature's connection).

• The Custom Proxy radio is hidden in the settings UI — you can no longer select it.
• If a chat still has a feature saved in proxy mode, DLE shows a boot-time deprecation notice and the AI call throws a descriptive error. /dle-health flags it with a fix: pick a Connection Profile.
• enableCorsProxy: true is no longer required for DLE's AI features. (ST's own AI requests may still need it for raw-URL backends.)
• Migration: open Connection → AI Connections, set the feature's Connection Mode to Connection Profile (or Inherit), and pick a saved Connection Manager profile. Any provider works.

Valid connection modes per feature are now: AI Search → profile; Session Scribe / Auto Lorebook → inherit / st / profile; AI Notepad / Librarian / Optimize Keys → inherit / profile.

Master toggle (About tab)

Setting	Default	Range	Description
Enable DeepLore Enhanced	Off	Toggle	Master switch (enabled). When off, no entries match, no AI calls run, no prompts get injected. The setting is mirrored in the bottom of the SillyTavern Extensions panel and on the About tab of the settings popup.

Vault connections (Connection tab → Obsidian sub-tab)

DLE supports multiple Obsidian vaults. Each vault has its own host, port, API key, HTTPS toggle, and enable flag. Entries from all enabled vaults merge into one index, disambiguated by vaultSource.

Field	Default	Description
Vault Name	(none)	Display name. Must match the actual Obsidian vault name exactly so deep links into Obsidian resolve.
Host	127.0.0.1	IP or hostname of the machine running Obsidian. Change for remote vault connections.
Port	27123	Port for the Obsidian Local REST API plugin. The plugin listens on 27123 for HTTP and 27124 for HTTPS by default.
HTTPS	Off	Toggle to use HTTPS. Requires a trusted certificate (the plugin's self-signed cert must be in your OS trust store, or you've imported it via the auto-diagnostic flow).
API Key	(none)	Bearer token from the Local REST API plugin's settings. Stored plaintext in ST's extension settings JSON.
Enabled	On	Toggle this vault on/off without deleting the connection.

Add Vault adds another connection. Test All verifies every enabled connection. Scan Vaults sweeps a port range looking for responding Local REST API instances.

Setting	Default	Range	Description
Multi-Vault Conflict Resolution	all	Dropdown	How to handle entries with the same title across vaults. all: keep all (disambiguated by vault source). first: keep the first vault's version. last: keep the last vault's version. merge: merge content from all vaults.

Vault tags (Connection tab → Obsidian sub-tab)

Setting	Default	Description
Lorebook Tag	lorebook	Obsidian tag (without #) that marks a note as a lorebook entry. Only notes with this tag get indexed.
Always-Send Tag	lorebook-always	Tag that forces a note to always inject regardless of keyword matches. Equivalent to constant: true.
Never-Insert Tag	lorebook-never	Tag that prevents a note from ever being injected, even on keyword match. Good for drafts or WIP notes.
Seed Tag	lorebook-seed	Tag for entries whose content is sent to the AI as story context on new chats. Not injected; only informs AI selection. See Features#New Chat Features.
Bootstrap Tag	lorebook-bootstrap	Tag for entries that force-inject when chat is short, then become regular entries. See Features#New Chat Features.
Librarian Guide Tag	lorebook-guide	Tag for entries Emma reads but the writing AI never sees. Used for writing-style guides. The conflict rule: if both lorebook-guide and a regular lorebook tag are set, guide wins at runtime.
New Chat Threshold	3	1-20

Field definitions path (Connection tab → Obsidian sub-tab)

Setting	Default	Description
Field Definitions Path	DeepLore/field-definitions.yaml	Vault path to the YAML file containing custom gating field definitions. Edit Fields opens the visual rule builder (add, remove, modify custom gating fields with type, operator, tolerance). The four default fields (era, location, scene_type, character_present) live in this file too.

Prompts (Prompts tab)

DLE's AI-facing default prompts (search, scribe, notepad, librarian, etc.) are editable. Override files live as .md in your vault, and a per-prompt grid lets you Export, Reload, and Reset each one. Two settings govern the tab:

Setting	Default	Description
AI Prompt Language (aiPromptLocale)	(follow UI)	Locale for the AI-facing default prompts, separate from the UI locale. Empty follows the SillyTavern UI locale. Pin to a specific language (e.g. English) if you do not trust machine-translated prompts to preserve model behavior. User-overridden prompt text is unaffected.
Prompts Folder Path (promptsFolderPath)	DeepLore/prompts/	Vault folder where editable prompt-override .md files live. Sibling to field-definitions.yaml.

AI connections (Connection tab → AI Connections sub-tab)

This sub-tab houses one accordion section per AI feature. Each section configures a connection channel: AI Search, Session Scribe, Auto Lorebook, AI Notepad (extract mode), Librarian, and Optimize Keys. The per-feature connection settings live with the feature in this reference, so look up each setting in its own section below.

Connection modes (see the v2.5 migration callout above — Custom Proxy is gone):

Mode	Meaning	Available on
profile	Use a saved Connection Manager profile. Any provider works.	All features
inherit	Reuse the AI Search connection.	All except AI Search (it is the inheritance root)
st	Use SillyTavern's own active connection.	Session Scribe, Auto Lorebook only

AI Search is the only feature that cannot inherit (it is the source). Librarian's default mode is inherit (it was profile before; changed in a settings migration). Each feature can override mode, profile, and model independently — even when inheriting, the per-feature Model / Max Tokens / Timeout still apply. The channels are deliberately independent so users can route Emma to a tool-calling model while AI search runs on something cheap.

Search mode (Matching tab)

Setting	Default	Description
Search Mode	two-stage	Dropdown. keywords-only: keywords only, no AI. two-stage: keywords pre-filter, then AI selects from the manifest. ai-only: full vault sent to AI. See AI Search.

Matching and budget (Matching tab)

Setting	Default	Range	Description
Messages to Search (Scan Depth)	4	0-100	Number of recent chat messages scanned for keyword matches. Set to 0 to disable keyword matching (AI search only).
Auto-match active character	Off	Toggle	Auto-match the active character's vault entry by name or keyword, even when not mentioned in chat. See Features#Active Character Boost.
Flexible Matching (BM25)	Off	Toggle	Supplement keyword matching with BM25/TF-IDF scoring. Catches partial or approximate keyword matches. The inverted index is built during index build.
Match Sensitivity (Fuzzy Min Score)	0.5	0.1-2.0	Shown when Flexible Matching is on. Minimum BM25 score for a fuzzy match. Lower is more permissive, higher is stricter.
Unlimited Entries	Off	Toggle	Remove the cap on how many entries get injected per generation.
Max Entries	10	1-100	Maximum entries to inject when Unlimited Entries is off. Sorted by priority.
Unlimited Token Budget	Off	Toggle	Remove the token budget cap. A warning toast fires if injected lore exceeds 20% of context.
Token Budget	3072	100-100000	Maximum total tokens to inject when Unlimited Token Budget is off. Entries are added in priority order until the budget hits.

Repetition control (Matching tab)

Setting	Default	Range	Description
Enable Repetition Control (Entry Decay)	Off	Toggle	Track entry freshness and adjust AI manifest priorities. Stale entries get a boost; frequently injected entries get a penalty. See Features#Entry Decay & Freshness.
Boost After N Skips	5	2-20	Consecutive generations an entry is skipped before its freshness boost kicks in.
Penalize After N in a Row	2	2-10	Consecutive generations an entry is injected before its frequency penalty kicks in.

Matching advanced (Matching tab → Show Advanced)

Setting	Default	Range	Description
Case Sensitive	Off	Toggle	When on, keyword matching respects case (Eris will not match eris).
Match Whole Words	Off	Toggle	When on, keywords use word boundaries (war will not match warning).
Keyword Occurrence Weighting	Off	Toggle	Entries with more keyword occurrences in the scan text sort higher within the same priority tier. May increase processing time on large vaults.
Chain Matching (Recursive Scanning)	Off	Toggle	After initial matches, scan matched entries' content for keywords that trigger more entries.
Max Recursion Steps	3	1-10	Maximum recursive scan passes.
Repeat Cooldown (Re-injection Cooldown)	0	0-50	Skip re-injecting an entry for N generations after last injection. 0 disables. Constants are exempt.
Skip Recently Injected (Strip Duplicate Injections)	On	Toggle	Skip re-injecting entries already injected in recent generations. Tracked per-chat. Constants are exempt.
Messages to check (Lookback Depth)	2	1-10	Number of previous generations to check for already-injected entries. Higher values dedupe more aggressively.
Filter Strictness (Contextual Gating Tolerance)	strict	Dropdown	How strictly contextual gating filters entries. strict: block when context dimensions are missing. moderate: block only on direct mismatch. lenient: allow when no context dimensions are set. Custom fields can override per-field.
Reverse priority order (priorityReversed)	Off	Toggle	Off (default): lower priority number wins, matching Obsidian/World Info convention. On: higher number wins. Affects budget allocation, the Librarian view, and the Cartographer display. The Browse popup's explicit priority sort is unaffected.
Compress imports by default (importCompressByDefault)	Off	Toggle	When on, World Info import compresses each entry's body (caveman-style) before writing to the vault, annotated compress: caveman in frontmatter. Per-import option can override.
Example Messages import handling (wiImportEmHandling)	append	Dropdown	How World Info import treats Example Messages entries (ST positions 5/6). append: convert to a normal entry with a "## Example Dialogue" subheader. skip: drop them.

Injection (Injection tab)

Setting	Default	Description
Injection Mode	extension	Radio. extension: classic setExtensionPrompt() with fixed position/depth/role. prompt_list: registers named prompts (deeplore_constants, deeplore_lore) that appear in SillyTavern's Prompt Manager list. Drag them wherever you want. Requires Chat Completion API.

The Injection Overview table on the Injection tab covers three injection types (Lore Entries, Author's Notebook, AI Notepad), each with its own position, depth, and role. Constants follow the Lore Entries position.

Setting	Default	Range	Description
Lore Entries: Position	In-chat	Dropdown	Where to inject lore. Options: Before Main Prompt, After Main Prompt, In-chat @ Depth.
Lore Entries: Depth	4	0-9999	Chat depth for in-chat injection (0 is the last message).
Lore Entries: Role	System	Dropdown	Message role for in-chat injection: System, User, or Assistant.
Author's Notebook: Position	In-chat	Dropdown	Same options as Lore Entries.
Author's Notebook: Depth	4	0-9999	Chat depth.
Author's Notebook: Role	System	Dropdown	System, User, or Assistant.
AI Notepad: Position	In-chat	Dropdown	Same options as Lore Entries.
AI Notepad: Depth	4	0-9999	Chat depth.
AI Notepad: Role	System	Dropdown	System, User, or Assistant.
Show Lore Sources Button	On	Toggle	Add a book icon to AI messages showing which entries were injected and why. See Features#Context Cartographer.
Injection Template	<{{title}}>\n{{content}}\n</{{title}}>	Text	Format for each injected entry. {{title}} is the entry name, {{content}} is the note body.
Allow World Info Scan	Off	Toggle	Let SillyTavern's built-in World Info system scan injected lore for WI keyword matches. Enables cross-system triggering.

Entries can override position, depth, and role via frontmatter. See Writing Vault Entries.

In prompt_list mode, the global Lore Entries position/depth/role settings are ignored. The Prompt Manager controls placement. Per-entry frontmatter overrides with custom depth or position still create separate injection groups that bypass the PM.

AI search (AI Search tab)

Visible when Search Mode is two-stage or ai-only.

Setting	Default	Range	Description
Connection Mode	profile	Radio	AI Search routes through a saved Connection Manager profile. (Custom Proxy mode was removed in v2.5 — see the migration callout at the top of this page.)
Connection Profile	(none)	Select	Pick a saved Connection Manager profile. Any provider works.
Model Override	(none)	Text	Optional model override. Leave empty to use the profile's model.
Max Response Tokens	1024	64-4096	Token limit for the AI response. Keep low; only a small JSON array is needed.
Timeout (ms)	20000	1000-999999	Maximum wait before falling back to keyword-only results. Local models may need 60000-120000ms; cloud APIs typically respond in 5000-15000ms. Cap is intentionally permissive — set this past 120000ms only if you know your provider routinely runs longer.
AI Confidence Threshold	low	Dropdown	Minimum confidence level for AI selections. low: accept all (high+medium+low). medium: accept medium and high. high: high only.
AI Error Fallback	keyword	Dropdown	What to inject when AI search errors out. keyword: fall back to keyword-matched results. constants_only: constants only. bootstrap_only: constants and bootstrap entries. none: nothing.
AI Empty Result Fallback	constants	Dropdown	What to inject when the AI intentionally returns []. constants: constants only. constants_bootstrap: constants and bootstrap entries. keyword: keyword results. none: nothing.

Test AI Search tests the connection. Preview AI Prompt shows the full prompt that would be sent. AI Stats shows session usage: AI calls, cache hits, estimated input/output tokens.

AI search prompt and context (AI Search → Show Prompt & Context)

Setting	Default	Range	Description
Messages Sent to AI (AI Scan Depth)	4	1-100	Number of recent messages sent to the AI for context. Can differ from the keyword scan depth.
Max Summary Length (chars) (Entry Description Length)	600	100-1000	Max characters for entry descriptions in the AI manifest. Only applies to entries without a summary field.
System Prompt Override	(none)	Text	Custom system prompt for AI selection. Empty uses the default. Supports the {{maxEntries}} placeholder.
Claude system prefix (Prepend "You are Claude Code")	Off	Toggle	Legacy proxy-mode toggle. Only had an effect in the removed Custom Proxy mode, so it is now a no-op in profile mode. Left in place for rollback.
Force merge system prompt into user message	Off	Toggle	Merge the system prompt into the user message. Use when the provider rejects or mishandles the system role (e.g., some Z.AI GLM versions). Try Prompt Post-Processing Semi/Strict first.
Use Session Notes as AI Context	Off	Toggle	Feed the Session Scribe's latest summary into AI search context for better entry selection. See Features#Use Session Notes as AI Context.
Custom Fields in AI Manifest (aiManifestIncludeFields)	(all)	Text	Comma-separated whitelist of custom-field names to send to the AI alongside each candidate entry. Empty includes every custom field that has a value. Populate (e.g. era, faction) to prune the manifest for token control.
Response Prefill Mode (responsePrefillMode)	off	Dropdown	Inject seed text as the last assistant message so the writing AI continues from it. off: no injection. anthropic-only: only when the model is Claude. all-providers: apply to every chat-completion source.
Response Prefill Seed (responsePrefillSeed)	(none)	Text	The prefill text injected when a prefill mode is active.

AI search filtering (AI Search → Show Filtering)

Setting	Default	Range	Description
Category Pre-filter (Hierarchical Pre-filter)	Off	Toggle	When on, large candidate sets (40+ keyword matches across 4+ categories) are narrowed by category in an extra AI call before final selection. When off, all keyword-matched entries go directly to the AI.
Category Filtering Strength (Hierarchical Aggressiveness)	0.8	0.0-0.8	How aggressively the category pre-filter culls entries. 0.0 keeps all categories. 0.8 trims up to 80%. The safety valve falls back to the full manifest if filtering would remove more than this fraction.
AI Entry Description Mode (Manifest Summary Mode)	prefer_summary	Dropdown	How entry descriptions are built for the AI manifest. prefer_summary: use summary if present, fall back to truncated content. summary_only: only include entries with summaries. content_only: always truncate content, ignore summaries.

Optimize Keys (AI Search → Show AI Utilities)

Settings for the /dle-optimize-keys command, which refines entry keywords using AI analysis.

Setting	Default	Range	Description
Optimize Keys Mode	keyword	Dropdown	keyword: refine existing keywords without AI. two-stage: use AI to analyze entry content and suggest better keywords.
Optimize Keys Prompt	(none)	Text	Override the default optimization prompt.
Connection Mode	inherit	Radio	inherit uses the AI Search connection, or pick profile (saved profile).
Connection Profile	(none)	Select	Profile mode only.
Model Override	(none)	Text	Override the model used for keyword optimization. Empty inherits from AI Search.
Max Tokens	1024	256-8192	Maximum tokens for the optimization response.
Timeout (ms)	30000	5000-999999	Request timeout for optimization calls.

Author's Notebook (Features tab → Author Notebook)

Setting	Default	Description
Enable Author's Notebook	Off	A persistent per-chat scratchpad injected into every generation. Edit via /dle-notebook or Open Author Notebook. See Features#Author's Notebook.

Position, depth, and role for the Author's Notebook live on the Injection tab. See the Injection table above.

AI Notepad (Features tab → AI Notebook)

AI-managed session notes that accumulate context across a conversation. The AI either writes notes inline using <dle-notes> tags (tag mode) or a post-generation extraction call pulls notes from the AI's response (extract mode). The accumulated notes get injected alongside regular lore. See Features#AI Notepad.

Setting	Default	Range	Description
Enable AI Notepad	Off	Toggle	Enable AI-managed session notes that persist across a conversation.
Capture Mode	tag	Radio	tag: AI uses <dle-notes> tags inline during generation. Free, no extra API calls. Best with flagship models. extract: a separate post-generation call extracts notes. Works with any writing model. Costs one extra API call per message.
Custom Tag Instruction	(none)	Text	Custom instruction prompt for tag mode. Empty uses the default.
Custom Extraction Prompt	(none)	Text	Custom extraction prompt for extract mode. Empty uses the default.
Connection Mode	inherit	Radio	Extract mode only. inherit uses AI Search, or pick profile.
Connection Profile	(none)	Select	Extract mode, profile only.
Model Override	(none)	Text	Extract mode. Override the extraction model.
Max Response Tokens	1024	256-8192	Extract mode. Maximum tokens for the extraction response.
Timeout (ms)	30000	5000-999999	Extract mode. Request timeout.

Position, depth, and role for the AI Notepad live on the Injection tab.

Session Scribe (Features tab → Session Scribe)

Setting	Default	Range	Description
Enable Session Scribe	Off	Toggle	Auto-summarize sessions to your Obsidian vault. See Features#Session Scribe.
Auto-Scribe Every N Messages	5	1-50	Number of new messages between automatic summaries. Tracked by chat position.
Session Folder	Sessions	Text	Vault folder where session notes get saved. Created if missing.
Messages to Include	20	5-100	Number of recent chat messages included as context for the summary.
Custom Summary Prompt	(none)	Text	Override the default summary prompt. The default covers events, character dynamics, revelations, and state changes.
Connection Mode	inherit	Radio	inherit (use AI Search), st (use SillyTavern's own active connection), or profile (saved profile).
Connection Profile	(none)	Select	Profile mode only.
Model Override	(none)	Text	Override the model used for summaries. Empty inherits from AI Search.
Max Response Tokens	1024	256-4096	Maximum tokens for the summary response.
Timeout (ms)	60000	5000-999999	Request timeout for summary generation.

Auto Lorebook (Features tab → Auto Lorebook)

Setting	Default	Range	Description
Enable Auto Lorebook	Off	Toggle	AI analyzes chat for entities not in the lorebook and suggests new entries with human review. See Features#Auto Lorebook.
Interval (messages)	10	3-50	Trigger auto-suggest every N messages.
Target Folder	(none)	Text	Obsidian folder for new entries. Empty saves to vault root.
Write directly (skip review popup)	Off	Toggle	When on, auto-suggested entries write to the vault immediately without the review popup. Use with caution.
Custom Prompt	(none)	Text	Override the default auto-suggest prompt.
Connection Mode	inherit	Radio	inherit (use AI Search), st (use SillyTavern's own active connection), or profile (saved profile).
Connection Profile	(none)	Select	Profile mode only.
Model Override	(none)	Text	Override the model used for suggestions. Empty inherits from AI Search.
Max Tokens	2048	256-4096	Maximum tokens for the suggestion response.
Timeout (ms)	30000	5000-999999	Request timeout for auto-suggest generation.

Use /dle-newlore to trigger on-demand at any time.

Librarian (Features tab → Librarian)

Tool-assisted lore retrieval and gap detection. When enabled, the writing AI can call search and flag tools during generation to look up vault entries and flag missing lore. The Librarian session UI (/dle-librarian) opens an interactive chat with Emma for drafting new entries. See Features#Librarian.

Tools

Setting	Default	Range	Description
Enable Librarian	Off	Toggle	Enables Librarian generation-time tools and the interactive session UI. Requires a tool-calling-capable connection; falls back to plain generation if unsupported.
Allow vault searches during generation (Search Tool)	On	Toggle	Enables the search tool during generation (gated by Enable Librarian).
Allow flagging missing lore (Flag Tool)	On	Toggle	Enables the flag tool during generation (gated by Enable Librarian).
Show tool call details on messages	On	Toggle	Show the "Consulted lore vault" dropdown on assistant messages when the AI used Librarian tools.
Per-message activity	Off	Toggle	When on, gap and flag records are tied to messages: cleared on new generation, kept on swipe, dropdown data persisted per-message. When off (default), gaps accumulate and dropdowns are ephemeral.
Searches per reply	2	1-10	Maximum search calls per generation.
Results per search	5	1-20	Maximum entries returned per search call.
Result size limit (tokens)	1500	500-5000	Token budget for search results returned to the AI per call.

Writing sessions

Setting	Default	Range	Description
Write Folder	(none)	Text	Destination folder in Obsidian for entries written from Librarian sessions. Empty saves to vault root.
Auto-draft when opening a gap (Auto-Send on Gap)	On	Toggle	Automatically send the draft prompt when opening a flagged gap in the Librarian session.
Connection Mode	inherit	Radio	inherit (use AI Search) or profile (saved profile). Defaults to inherit. Librarian intentionally has its own channel so users can route Emma to a different model.
Connection Profile	(none)	Select	Profile mode only.
Model Override	(none)	Text	Override the model. Empty inherits from AI Search (when in inherit mode).
Session Max Tokens	4096	1024-16384	Maximum tokens for Librarian session responses.
Session Timeout (ms)	120000	10000-999999	Request timeout for Librarian session AI calls. The 120s default leaves headroom for forced-final-response loops on large contexts and reasoning models.

Writing sessions advanced

Setting	Default	Range	Description
Manifest Budget (chars)	8000	0-200000 (UI: 2000-30000)	Maximum characters for the vault manifest in Librarian session prompts. Char-based truncation; divide by ~3.5 for a token estimate.
Related Entries Budget (chars)	4000	0-200000 (UI: 1000-10000)	Maximum characters for related-entries context.
Chat Context Budget (chars)	4000	0-200000 (UI: 1000-10000)	Maximum characters for chat context.
Draft Budget (chars)	4000	0-200000 (UI: 2000-10000)	Maximum characters for the serialized draft JSON.
System Prompt Mode	default	Radio	default: built-in Librarian system prompt. append: add custom text after the default. override: replace the role/persona section only (manifest, gap context, chat, draft, tools, response format still appended). strict-override: pure passthrough, custom text IS the entire system prompt.
Custom System Prompt	(none)	Text	Custom system prompt text. Used in append, override, and strict-override modes.

Graph (Features tab → Graph)

Settings for the interactive entry relationship graph (/dle-graph). The graph also has its own in-window settings panel (gear icon) with live sliders and presets. Changes save to your DLE settings and persist.

Setting	Default	Range	Description
Default Color Mode	type	Dropdown	Initial node coloring scheme. Options: type (by entry type), priority, centrality, frequency, community (Louvain clusters).
Hover Dim Distance	3	0-15	BFS hops kept visible on hover. Nodes farther than this get dimmed. 0 disables.
Focus Tree Depth	2	1-15	Hop depth for ego-centric focus mode. Controls how many relationship hops show from the focused node.
Show Labels	On	Toggle	Show node title labels on the graph.
Repulsion	0.3	0.1-5.0	ForceAtlas2 repulsion coefficient. Higher pushes nodes further apart.
Gravity	11.0	0.1-20	Pull toward center. Higher keeps the graph more compact.
Damping	0.50	0.3-0.98	Velocity damping. Higher settles faster.
Hover Falloff	0.55	0.3-0.85	Transmission per hop for hover dimming. Light propagates as t^d (mirrors-and-lasers model). Higher lets light reach further hops; lower dims more aggressively.
Hover Ambient	0.06	0.0-0.2	Ambient floor for off-set elements on hover. Nodes beyond the dim distance never go below this opacity.
Edge Filter Alpha	0.05	0.01-0.5	Serrano disparity filter threshold. Lower shows fewer (more significant) edges.
Node Size Mode	centrality	Dropdown	How node sizes are determined. centrality (size by graph centrality), priority (by entry priority), uniform (all same).

Graph physics presets (Compact, Balanced, Spacious, Ginormous) set Repulsion / Gravity / Damping to recommended combinations. Individual sliders fine-tune from there.

Index and cache (System tab → Show Advanced)

Setting	Default	Range	Description
Cache Duration (seconds)	300	0-86400	How long to cache the vault index before re-fetching. 0 always fetches fresh (slower).
Auto-Sync Interval	0	0-3600	Seconds between automatic vault re-checks. 0 disables. See Features#Vault Change Detection & Auto-Sync.
Show Sync Change Toasts	On	Toggle	Show toast notifications when vault changes get detected during index refresh.
Index Rebuild Trigger	ttl	Dropdown	When to rebuild the vault index. ttl: rebuild when cache duration expires. generation: rebuild every N generations. manual: only rebuild on explicit refresh.
Rebuild Every N Generations	10	1-100	Shown when trigger is generation. How many generations between automatic index rebuilds.

Refresh Index clears the cache and re-fetches all entries. Test Match simulates a generation to show which entries would match. Browse opens the entry browser.

System advanced (System tab)

Setting	Default	Range	Description
Review Response Tokens	0	0-100000	Token limit for /dle-review responses. 0 is auto (uses your connection profile's Max Response Length).
Debug Mode	Off	Toggle	Log detailed match info to browser console (F12). Shows keyword matches, AI results, gating, token counts, injection details.

Internal and migration settings

These keys appear in defaultSettings but are not surfaced in the UI as user-editable. Listed here for completeness (power users editing the JSON or reading diagnostics will see them).

Key	Default	Description
lenientAuthoring	true	Frontmatter parser auto-fixes common authoring mistakes (case-mismatched field names, quoted numbers, comma-string keys) and records a warning. Set to false to match pre-v2 strict behavior (silent drops).
drawerPinned	false	Drawer UI state. Whether the drawer is pinned open.
drawerCompactTabs	false	Drawer tabs render as icon-only (true) or icon plus text (false).
advancedVisible	{}	Per-section "Show Advanced" toggle states.
promptPresets	{}	Saved prompt presets keyed by tool name ({ [toolKey]: { [presetName]: promptText } }).
analyticsData	{}	All-time analytics counters.
_wizardCompleted	false	First-run setup wizard completion flag.
settingsVersion	5	Settings schema version. Migrations run on load when stored version is behind.
graphSavedLayout	null	Saved graph node positions: { positions: { title: { x, y } }, timestamp }.
aiSearchEnabled	false	Internal mirror of aiSearchMode being non-keywords-only. Auto-managed; do not edit.

Automatic features (no settings)

These work automatically with no configuration:

• Obsidian circuit breaker: the Obsidian connection uses a per-vault circuit breaker (closed/open/half-open) with exponential backoff (2s to 15s). Keyed by host:port so each vault has independent failure tracking. Resets when a call succeeds.
• AI circuit breaker: AI search has its own circuit breaker (2 consecutive failures to trip, 30s cooldown). Prevents repeated full-timeout waits when the AI service is down. The AI throttle (500ms minimum between calls) does not trip the breaker. User aborts, timeouts, rate-limit (429), and auth errors also do not trip the breaker.
• IndexedDB persistent cache: the parsed vault index is saved to IndexedDB (DeepLoreEnhanced database, vaultCache store) after every successful build. On page load, hydrates instantly from cache, then validates in background.
• Reuse sync: auto-sync fetches all file contents but skips re-parsing/tokenizing unchanged entries (detected by content hash). Falls back to full rebuild automatically if reuse fails.
• Hierarchical manifest clustering: for vaults with 40+ selectable entries and 4+ categories, the optional Category Pre-filter uses a two-call AI approach for better scaling. Off by default.
• Sliding window AI cache: AI search cache tracks manifest and chat hashes separately for smarter cache reuse. New chat lines are reused unless they reference vault entities.
• Confidence-gated budget: AI search over-requests entries (2x), sorts by confidence tier before applying the budget cap.
• Confidence tiers: AI selections carry a confidence level (high/medium/low); the AI Confidence Threshold setting decides which tiers are accepted before the budget cap applies.
• Settings migrations: run automatically on load when the stored settingsVersion is behind the current schema (5). 1→2 consolidated Librarian model fields. 2→3 changed Librarian's default connection mode from profile to inherit for unconfigured users (explicit profile selections preserved). 3→4 (v2.5) flipped any feature still on Custom Proxy mode to profile and shows a one-time deprecation notice — see the migration callout at the top of this page. 4→5 added the wiImportEmHandling default.