Search Tools

Search Tools provides AI characters with web search capabilities, allowing characters to retrieve real-time information during conversations and inject search results into the context. The search plugin supports multiple search engine backends and provides two working modes to accommodate different use cases.

Two Working Modes

Tool Mode

Search functionality is registered as a callable tool for the creative LLM in the Function Call Runtime. When the AI determines it needs to search for information, it proactively initiates a tool call.

• Enabled via the enabled configuration option
• Registers luker_web_search and luker_web_visit as function tools
• AI autonomously decides when to search and what to search for
• Search results are injected into the conversation context as tool call return values

Pre-request Agent Mode

Before the creative LLM generates a response, an independent search Agent runs automatically. The Agent analyzes the current conversation content, determines whether a search is needed, and writes search results into World Info entries.

• Enabled via the preRequestEnabled configuration option
• The Agent uses independent LLM presets (agentPresetName) and API presets (agentApiPresetName), which can use different models from the creative LLM
• Agent maximum execution rounds are controlled by agentMaxRounds
• Search results are automatically created as World Info entries that the creative LLM can directly read during generation
• Supports stopping Agent execution via the Stop button on toast notifications

Choosing Between Modes

Tool Mode is suitable for scenarios where the model needs to autonomously decide when to search during conversation — for example, when a user asks "What's in the news lately?" the model will automatically call the search tool to get information. Search behavior is autonomously triggered by the model based on conversation content.

Pre-request Agent Mode automatically executes searches before each AI generation, with the search process completely separated from the main conversation. Suitable for scenarios where you want every response to be backed by the latest information.

Supported Search Engines

Search Engine	Description	Configuration Requirements
DuckDuckGo	Default search engine, no API key required	Works out of the box
SearXNG	Self-hosted meta search engine, privacy-friendly	Requires self-hosted instance URL
Brave Search	Search API provided by Brave	Requires API key

The search engine is selected via the provider configuration option, with each engine's independent configuration stored in providerSettings. You can also configure the safe search level via safeSearch.

World Info Entry Management

In Pre-request Agent Mode, search results are automatically created as World Info entries. The behavior of these entries can be controlled through the following configuration options:

Setting	Description
lorebookDepth	Injection depth of entries in the prompt
lorebookRole	Role marker for entries (e.g., system, assistant)
lorebookEntryOrder	Sort weight of entries
lorebookPosition	Injection position of entries

Search result entries can be set to constant (always active) or activated through keyword matching, depending on the Agent's judgment and configuration.

Global API

The search plugin provides a global API for integration by other plugins. For example, the Character Card Editor has integrated search capabilities, enabling web searches for reference materials in Studio to assist with character card editing.

Properties

Property	Type	Description
toolNames	Object	Tool name constants object, containing SEARCH and VISIT

Methods

Method	Return Value	Description
getToolDefs()	Array	Returns the search tool definition array, including tool names, parameter schemas, etc.
isToolName(name)	boolean	Determines whether a given name is a tool name registered by the search plugin
invoke(call, options)	Promise	Invokes a search tool; call contains the tool name and parameters
search(args)	Promise	Directly executes a search and returns results
visit(args)	Promise	Directly visits a specified web page and returns page content
getSettings()	Object	Returns a snapshot of current search plugin settings

Usage Example

const api = Luker.searchTools;
if (api) {
  // Check if it's a search tool
  api.isToolName('luker_web_search'); // true

  // Get tool definitions
  const defs = api.getToolDefs();

  // Directly call search
  const results = await api.search({ query: '...', maxResults: 5 });

  // Visit a web page
  const page = await api.visit({ url: 'https://...', maxChars: 10000 });
}

Result Reuse

The search plugin implements a fine-grained result reuse mechanism to avoid redundant searches in unnecessary scenarios.

The system automatically determines whether the previous search Agent's results can be reused. Reuse requires all of the following conditions to be met:

1. The current chat key matches the previous snapshot's chat key
2. The user anchor (characteristics of the last user message) matches the previous snapshot
3. The current generation type belongs to a reusable type (e.g., regeneration, continue generation scenarios)

When results are reused, the UI displays a (reused) marker, indicating that cached search results were used for this generation.

Impact of Message Operations

The search plugin monitors message deletion and editing events, automatically maintaining internal state consistency:

• Message deletion — Updates search state based on the range of deleted messages, ensuring no references to search results from deleted messages
• Message editing — Updates search state based on the edited message position, potentially invalidating cached search results

Configuration Reference

Setting	Description
Enable Search Tools	Enable Tool Mode, allowing the model to autonomously call search during conversation
Enable Pre-request Agent	Enable Pre-request Agent Mode, automatically searching before each generation
Search Engine	Select search engine (DuckDuckGo / SearXNG / Brave Search)
Search Result Count	Number of results returned per search
Page Extract Characters	Maximum text length extracted when visiting web pages
Safe Search	Safety filtering level for search results
Agent API Preset	API connection preset for the pre-request Agent (empty uses main connection)
Agent Preset	Preset for the pre-request Agent (empty uses main preset)
Agent Max Rounds	Maximum search rounds for the pre-request Agent

Related Pages

• Character Card Editor — Web search capabilities in Studio
• CardApp — Character card application concept