Agentic Lorebook

An Agentic Lorebook lets the Director choose lorebook entries by reading the scene. It can find relevant lore even when none of the entry's keywords appear.

How it works

A World contains lorebook entries. An entry can be active because it is:

• Constant: always included in the character context.
• Keyword-activated: included when its keywords match recent messages, while Agentic Lorebook is off.
• Selected by the Agent: chosen by the Director for the current scene.

When Agentic Lorebook is on, the Director has full control over non-constant entries. Keyword activation is disabled, so only the Director's picks are included. An empty selection or a failed selection call includes no non-constant entries. Constant entries stay active.

What the agent sees

How does the agent decide which entries are relevant? These info will be sent to it:

• The lorebook's name
• The entries' names
• Each entry's keywords as relevance hints (up to three, excluding the entry name)

Constant entries are excluded from the catalog. They stay in the system prompt that every Agent call shares, selection included.

Enable it

Open Settings → Agents and turn on Agentic Lorebook. The global Agent toggle must also be on.

The Director receives a short catalog of non-constant entries and selects the ones that fit the current scene. This uses one additional lightweight model call per turn. If there are no selectable entries, no selection call is needed. Turning off Agentic Lorebook or the global Agent restores normal keyword activation.

See Lorebooks for entry types, triggers, macros, and import rules.