Web Search - Cloosphere Guide

Admin › Settings › Web Search

Configure the AI’s web search functionality.

With web search enabled, agents can reference real-time internet information when answering.

General

Setting	Description	Default
Web Search	Web search feature ON/OFF (master toggle under the General section)	Disabled

Supported web search engines.

Engine	API Key Required	Notes
SearXNG	Not required (self-hosted)	Privacy-focused metasearch engine. Requires Query URL
Google PSE	Required	Google Programmable Search Engine. Requires API Key + Engine ID
Brave Search	Required	Brave Search API
Kagi	Required	Premium search engine
Mojeek	Required	Independent crawler-based search engine
Bocha	Required	AI search API
Serpstack	Required	Google search results API
Serper	Required	Google search results API
Serply	Required	Search results API
SearchAPI	Required	Multi-engine search API. Engine selectable
SerpAPI	Required	Google search results scraping. Engine selectable
DuckDuckGo	Not required	Free, privacy-focused
Tavily	Required	AI-optimized search API
Jina	Required	AI search API
Bing	Required	Microsoft Bing Search V7. Requires Endpoint + Subscription Key
Exa	Required	Semantic search API
Perplexity	Required	AI-based search
Sougou	Required	Chinese search engine. Requires API sID + SK

The detail settings below are only visible when Web Search is on.

Setting	Description	Default
Number of results	Number of search results to fetch	`5`
Concurrent requests	Number of parallel search requests	`1`
Domain filter list	Allow/block domain list (comma-separated)	-
Bypass Embedding and Retrieval	Bypass the embedding and vector search steps for web search results and pass the raw content directly to the context	Disabled
Trust proxy environment	Trust and use the system proxy environment variables when making requests	Disabled

Tuning guide

Number of results — Raising it gathers information more broadly and lowers the chance of missing relevant content, but increases context tokens and cost, mixes in irrelevant pages, and slows responses. Lowering it references only the essentials quickly but may miss needed information.
Concurrent requests — Raising it fetches pages in parallel for faster search, but raises the risk of hitting the search API’s or target sites’ rate limits or being blocked. Lowering it is more stable but collects results more slowly.

The engine that fetches web page content from search result URLs.

Engine	Strengths
Default	Simple HTTP request to extract text. Toggle Verify SSL Certificate to set SSL certificate verification
Playwright	Supports JavaScript rendering for SPA pages. Set WebSocket URL and Timeout
Firecrawl	Advanced web crawling and content refinement. Requires API Base URL and API Key
Tavily	AI-optimized content extraction. Set Extract Depth

For pages heavy with JavaScript rendering, Playwright or Firecrawl is recommended. The default loader only processes static HTML.

Use YouTube video subtitles as search sources.

Setting	Description
Languages	Subtitle languages to fetch (comma-separated, e.g., `ko,en`)
Proxy URL	Proxy URL to use when accessing YouTube (e.g., `https://user:password@host:port`)