Skip to content

First Steps

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

First steps

You've installed DeepLore and connected your vault (Quick Start). This page walks through building a useful starter vault, picking the right pipeline mode, and tuning matching for your workflow.

Pick a pipeline mode

DLE has three modes. The setup wizard sets one for you; you can change it any time in Settings → Matching.

Mode How it works Cost Best for
Keywords only Scans recent chat messages for keywords from your entries Free Small vaults (under 50 entries), no AI provider configured
Two-stage Keywords pre-filter, then AI search ranks the candidates against summaries ~1 extra provider call per turn The default. Medium and large vaults (50+ entries)
AI only AI search reads the entire vault manifest each turn Higher: full manifest tokens per call Small vaults where you want context-only matching, no keywords needed

Two-stage is the recommended default. Keyword-only is fine while you're learning the system; switch to two-stage once you have ~30+ entries and the keyword cast net starts missing things.

Write your first entries

A useful starter vault has 3 to 5 entries covering the kinds of things your story leans on. Examples:

A character entry

---
tags: [lorebook]
keys: [Eris, the Spymaster]
priority: 20
summary: "Eris, the protagonist's spymaster and closest enforcer. Select when espionage, intelligence, interrogation, or loyalty comes up."
---
  • Priority 20 marks an inner-circle character (lower number = higher priority)
  • Keys cover both the name and the role title

A location entry

---
tags: [lorebook]
keys: [The Crimson Den, tavern, underground bar]
priority: 50
summary: "An underground tavern serving as a neutral meeting ground. Select when taverns, drinking, underground dealings, or nightlife come up."
---
  • Priority 50 is the standard middle priority
  • Keys cover both the proper name and generic terms that should trigger it

A lore-concept entry

---
tags: [lorebook]
keys: [bloodbond, blood bond, feeding dependency]
priority: 35
summary: "The biological dependency created when a vampire feeds from a mortal. Select when feeding, biting, addiction, or chattel dynamics come up."
---
  • Priority 35 for core worldbuilding
  • Multiple keyword variants catch different phrasings

Constants, bootstrap, and seed entries

Tag combinations control when entries fire:

  • Regular entries (just the lorebook tag): inject only when keywords match (or when AI search picks them in two-stage mode)
  • Constants (add lorebook-always): always inject on every generation. Use for core world rules, persistent character sheets, writing instructions
  • Bootstrap entries (add lorebook-bootstrap): force-inject when chat is short (first few messages), then revert to regular matching once chat history grows
  • Seed entries (add lorebook-seed): sent to AI search as story context on new chats, never injected into the writing AI

Keep constants minimal. Every constant spends context budget on every generation.

Tune scan depth and budget

Scan depth

How many recent chat messages DLE scans for keywords:

  • 2 to 3: matches recent conversation only. Responsive but narrow
  • 4 to 6: the default. Good balance
  • 8 to 10: broader matching, may pull in less relevant entries
  • 0: disables keyword scanning entirely. Use only with AI-only mode

Token budget

Maximum tokens DLE can inject per generation:

  • 1000 to 2000: conservative, good for smaller contexts
  • 3000 to 5000: balanced, works for most setups
  • Unlimited: DLE uses as much as it needs. Watch your overall context usage

Start with scan depth 4 and budget 3000, then adjust based on /dle-inspect output.

Enable AI search

When you're ready to switch from keywords-only to two-stage:

  1. Go to Settings → Matching → Search Mode and select Two-Stage
  2. In Settings → AI Search, pick a connection mode:
    • Profile: uses a SillyTavern Connection Manager profile (recommended)
    • Proxy: routes through SillyTavern's CORS proxy to a custom endpoint (advanced)
  3. Select or create a connection profile pointed at a fast, cheap model (Haiku, GPT-4o-mini, or any local model that handles structured output well)
  4. Run /dle-status. It should show "AI search: enabled"

AI search earns its keep when your vault is large or entries have nuanced triggers that keywords can't capture. Example from the README: a scene about "the consequences of breaking an oath" pulls in your Bloodchain entry without the word "Bloodchain" appearing.

Audit entry quality

Run /dle-health regularly. It checks for:

  • Missing or empty keywords
  • Duplicate keywords across entries
  • Missing summaries (AI search needs them)
  • Oversized entries (may dominate your budget)
  • Broken wikilinks
  • Configuration issues

Aim for grade A or A+.

Verify what's injecting

Three diagnostics show what the pipeline picked:

  1. /dle-inspect: shows the last pipeline trace. What matched, why, and what was injected
  2. Context Cartographer (book icon on AI messages): shows the per-message source list with token counts
  3. /dle-simulate: replays the entire chat showing which entries activate at each message

Common early mistakes

Mistake Fix
Too many constants Keep constants under 5 to 10. Use keywords for everything else
Keywords too generic "the", "and", "said" match everything. Use specific terms
Keywords too specific "Lord Vexathorn the Undying" won't match "Vexathorn". Add aliases
No summaries AI search needs summaries to work. Write one for every entry
Oversized entries (5000+ tokens) Split into smaller focused entries or use scan-depth overrides
Every entry at priority 50 Use the full range: 20 (critical), 35 (core), 50 (standard), 60+ (flavor)

Next steps

DeepLore Wiki

Getting started

Features

Reference

Help

Clone this wiki locally