Introduction

Internet connectivity features are divided into two main components: search engines and web crawling tools. Search engines will search the entire web for pages related to user queries, while web crawling tools can examine the content of specific webpages. You can add relevant web search logic to the system prompts to enhance the stability of agent searches.

Search Engines

• duckduckgo (completely free, not accessible from within China)
• searxng (can be deployed locally via docker)
• tavily (requires API key)
• bing (requires API key)
• google (requires API key)
• brave (requires API key)
• exa (requires API key)
• serper (requires API key)
• Bocha (requires API key)

Web Crawling

• jina (can be used without API key, for web crawling)
• crawl4ai (can be deployed locally via docker, for web crawling).

Other Configurations

1. Internet Connection Timing: Similar to knowledge base, you can choose before thinking, after thinking, or both

• Before Thinking: Search the internet before the large model sees user input, then return all internet results along with the original question to the model
• After Thinking: The large model determines whether to search the internet based on the knowledge base description and reformulates questions for more precise queries

1. Number of Results: Determines the number of results returned from internet searches