Skip to content

AI Powered Tools

Jump to bottom
pixelnull edited this page Apr 25, 2026 · 5 revisions

AI-Powered Tools

Features that use AI to help you build, maintain, and grow your vault. These run alongside AI Search (which selects entries during generation); the tools on this page actively create or improve vault content.

The headline v2 feature on this page is the Librarian, which has two halves: writing-AI tools (search and flag) that fire mid-generation, and Emma, a separate chat agent that helps you author the missing entries.

Note

Each tool below has its own independent connection channel. You can route Emma to one model, the Scribe to another, AI Search to a third. Defaults inherit from AI Search.

The Librarian (Emma)

The Librarian is how your vault grows as you roleplay. Two halves, one feature.

Half 1: writing-AI tools. During generation, the writing AI gets two function-calling tools: search and flag. When it reaches for a detail and isn't sure if your vault has it, it calls search. If the vault genuinely doesn't have it, it calls flag instead of inventing.

Half 2: Emma. A separate chat agent (her own connection channel, her own toolset) for authoring entries. You open a flag from the Librarian inbox in the drawer, and chat with Emma about the gap. She checks what already exists, finds similar entries to dedupe against, pulls in any style guides marked lorebook-guide, and drafts the new entry. Write-to-vault saves it back to Obsidian.

Important

The Librarian auto-enables function calling on the active connection. If you disable function calling elsewhere on that profile, tool invocations break mid-generation with no error toast.

Writing-AI tools (search and flag)

Both fire inside the writing AI's normal generation. Tool activity collapses into one expandable dropdown on the final assistant message; you still get one clean reply per turn.

Tool Args What it does
search queries: string[] (up to 4) BM25 vault search. Returns the single best hit with full content, plus up to 3 graph-linked entries in summary form. Excludes already-injected entries and lorebook-guide entries.
flag title, reason, urgency (low/medium/high), flag_type (gap/update), entry_title Records a gap or update request to the Librarian inbox. Re-flagging a similar gap merges with the existing record (frequency increment, urgency escalation).

The writing AI is capped at librarianMaxSearches searches per generation (default 2) and 5 flag calls. Internally a third tool, write, captures the final prose and transitions the loop from SEARCH to FLAG phase.

Emma's toolset

Inside an Emma session, she has her own internal tool set (12 tools). Highlights:

Tool Purpose
search_vault BM25 search across the vault
get_entry Truncated preview + metadata
get_full_content Full entry content; populates her draft state
find_similar Duplicate detection before authoring a new entry
get_links / get_backlinks Walk resolvedLinks outgoing or incoming
list_entries Filter by type and tag
get_recent_chat Last N messages from the active chat
compare_entry_to_chat Side-by-side entry + recent chat
flag_entry_update Write a gap record from inside the session
get_writing_guide Fetch entries tagged lorebook-guide. Available only when guide entries exist.

The lorebook-guide tag marks an entry as Librarian-only. It is never injected into the writing AI through any pipeline path; only Emma can see it via get_writing_guide.

Connection channel

The Librarian has its own connection channel (librarianConnectionMode), independent of AI Search. Defaults to inherit (resolves to AI Search settings); set it to profile or proxy to route Emma through a different model. Tool calling is required, so the resolved provider must support function calling.

Supported tool-calling providers: Claude, Gemini (makersuite/vertexai), OpenAI-compatible, Cohere. Sources without tool support (ai21, perplexity, nanogpt, pollinations, moonshot) cannot drive the Librarian; the writing AI falls through to ST's normal generation when tool calling is unsupported, and a deduplicated warning fires.

Setup

  1. Enable Librarian in Settings → Features → Librarian
  2. Pick a connection mode (default inherit works if your AI Search profile is tool-calling)
  3. Verify with /dle-health (checks tool-calling support on the resolved profile)
  4. Roleplay normally; gaps will accumulate in the Librarian inbox in the drawer

See also: librarianPerMessageActivity (default off) which changes whether gaps clear on each new generation. Off is the original behavior (gaps accumulate across messages); on ties gap and dropdown lifecycle to per-message scope.

Librarian tab with a flagged worldbuilding gap opened

Session Scribe

Auto-summarizes your roleplay into timestamped markdown notes written back to your Obsidian vault. The next summary feeds the prior summary back as context, so notes build on each other instead of repeating.

Flow:

  1. After every N new messages (default 5), Scribe fires from the CHARACTER_MESSAGE_RENDERED handler
  2. It summarizes the recent message window using the configured prompt
  3. The summary is written as a markdown file with frontmatter (tags: lorebook-session, date, character) to the configured Session Folder (default Sessions)
  4. The vault re-indexes so the new note is searchable

On-demand: /dle-scribe writes a summary immediately. Optional focus topic: /dle-scribe What happened during the trial?

Browse past summaries: /dle-scribe-history opens a popup of the current chat's notes.

Connection options:

  • Inherit (default): resolves to AI Search settings
  • Connection Profile: any saved Connection Manager profile. Use this to route summaries through a different (often cheaper) model.
  • Custom Proxy: a separate Anthropic-compatible Messages API endpoint (e.g., claude-code-proxy)

Setup:

  1. Enable Session Scribe in Settings → Features → Session Scribe
  2. Set the auto-scribe interval (every N messages)
  3. Set the Session Folder
  4. Pick a connection mode
  5. Optionally customize the summary prompt and message window depth (scribeScanDepth, default 20)

Defaults to know: message window 20 messages, response token limit 1024, timeout 60s.

Session Scribe output in Obsidian showing the vault folder structure on the left and a timestamped session note with Properties metadata, Key Events and Plot Developments bullet points, and a Character Dynamics section

Auto Lorebook

Scans recent chat for characters, places, items, and concepts that lack a vault entry, then proposes new entries you can review and accept.

Flow:

  1. Every N messages (default 10), or on-demand via /dle-newlore, the AI scans recent chat
  2. It compares against existing entries and identifies gaps (case-insensitive title match)
  3. Suggestions appear in a popup with title, type, keywords, summary, and content
  4. Accept to write the entry to Obsidian, or reject to skip

Connection options: Inherit (default), Connection Profile, or Custom Proxy.

Setup:

  1. Enable Auto Lorebook in Settings → Features → Auto Lorebook
  2. Set the trigger interval
  3. Optionally set a target folder for new entries
  4. Pick a connection mode

/dle-newlore triggers on-demand without enabling automatic suggestions. Accepted entries are written with proper frontmatter (type, priority, tags, keys, summary).

Optimize Keys

/dle-optimize-keys <entry name> asks an AI to analyze an entry and suggest better keywords. The AI considers the entry's content, summary, and current keys to recommend additions and refinements: synonyms, related terms, trigger phrases.

When to use: an entry isn't triggering as expected, or you want a second opinion on key choices after writing the entry.

Modes (optimizeKeysMode):

  • keyword: pure keyword optimization without AI
  • two-stage: AI analyzes content and suggests keys

Connection: independent (Inherit / Profile / Proxy). Defaults to Inherit (resolves to AI Search settings).

Suggestions appear in a popup for review. Apply changes in Obsidian, then run /dle-refresh to pick them up.

Auto-Summary Generation

/dle-summarize scans every indexed entry, finds those without a summary frontmatter field, and generates summaries one at a time. Each suggestion is shown in a review popup where you can edit, approve, or skip before it is written to the entry's frontmatter in Obsidian.

When to use: after importing entries from World Info, or when you have a backlog of entries without summaries. The summary field is what AI Search reads to decide whether an entry is relevant, so filling missing summaries directly improves selection quality.

Scribe-Informed Retrieval

When enabled, the Session Scribe's latest summary is fed into the AI Search context as additional story background. This widens what AI Search "knows" about the ongoing narrative beyond the scan-depth window, so it can pick entries relevant to long-running plot arcs.

Setup: enable Scribe-Informed Retrieval in Settings Reference.

Requires:

  • Session Scribe enabled
  • At least one summary written for the current chat

Most useful for long conversations where important plot points have scrolled past the AI Scan Depth window.

Related pages

DeepLore Wiki

Getting started

Features

Reference

Help

Clone this wiki locally