Web search
Web search and page fetching for the agent, with multiple search backends.
Let the agent search the web and fetch pages.
Two tools
- web_search — keyword search. Parameters:
query(required),max_results(optional, default 10, range 1–20). 60-second timeout. - web_fetch — fetch the main content of a URL. Parameters:
url(required),max_length(optional, default 8000),extract_links(optional, default false). Waits up to 30 seconds per URL.
Models from Claude, OpenAI, and Kimi ship with built-in web search — it works out of the box. With other models, pick a search backend below (Bocha / Serper / custom) and set a key first; only then does the agent get search capability.
CLI command /websearch
| Subcommand | What it does |
|---|---|
status | Show the current configuration |
on / off | Enable / disable web search |
set-engine | Pick a backend: auto / bocha / serper / custom |
set-url | Set a custom backend URL |
set-key | Set the backend API key |
Desktop
You can also toggle web search, pick a backend, and set the key in Settings · Agent · Web search (turn on "Show advanced settings" at the bottom of the settings sidebar first).
Common patterns
- "Look up the latest usage of X" → the agent finds it with web_search, then reads the page with web_fetch
- In mainland China, set the backend to
bocha— no extra network configuration needed