Settings: Web search
Choose the search service, result size, web fetch, research mode, fact-check and workspace memory.
Settings → Web search – "Pick a service the agent can use for web search."
Search services#
Tap Add Provider and search the list: Built-in (DuckDuckGo, default, no key), CX Agent, Bing, Tavily, Exa, Brave, Perplexity, Serper, Jina (search and scrape URLs), Firecrawl, SearXNG (your instance URL, engines, language, login), LinkUp, Grok, Tinyfish, Ollama, Bocha, Metaso, Zhipu, Doubao, and Custom JS (write your own search and optional scrape script). Each has its own fields (API key, depth, max tokens, summary…) and a Test Search box. Each row can be edited, deleted and selected for Search and Scrape.
Common options#
| Option | Meaning |
|---|---|
| Result Size | How many results per search. |
| Web fetch and extract | Let the assistant fetch and read pages directly (on by default). Off = rely only on the search provider. |
| Web Research Mode | Use multiple sources and require citations in research answers. |
| Fact-check & Grounding | Verify important claims with multiple sources and show a confidence score. |
| Project Workspace Memory | Include the active workspace context and project files. |
Per chat#
Add to task → Web search: Off, Local (this service) or Model (built-in model search).