Web Module

Module id web. Search the public web and fetch pages. Default search backend is DuckDuckGo (no API key). Optional fallbacks: Brave, Tavily, SearXNG, Google.

There is no separate web.extract or web.download tool. Main-content extraction is web.fetch with extract: true. For a persistent interactive browser session, prefer the browser module.

Module declaration

yaml

tools:
  modules:
    web:
      config:
        backend: duckduckgo
        # fallback: brave
        # allowed_domains: []
        # blocked_domains: []
  capabilities:
    default_policy: auto
    grant:
      - module: web
        actions: [search, fetch]

Tools

Action	Chat UI label	Purpose
web.search	Search	Search via the configured backend.
web.fetch	Fetch	Open / fetch a URL (optional extract, actions, screenshot, crawl, profile, live).

The LLM always calls the real FQN (web__search, web__fetch) - the chat UI label is a display-only transform, not a callable alias.

web.search

Param	Notes
query	Required.
limit	Max results (default 10).
fetch_content	Inline main page text per hit (default true).
time_range	day / week / month / year.

web.fetch

Param	Notes
url	URL to open; omit to re-read the current page in-session.
format	text / markdown / html.
extract	Main article body only.
prompt	Focus extraction on a topic.
actions	Page interactions before reading.
screenshot / live / profile / crawl	See tool schema on a running daemon.

Search backends

Backend	API key	Notes
duckduckgo (default)	no	Free default.
brave	yes	Production-quality.
tavily	yes	Agent-oriented snippets.
searxng	instance URL	Self-hosted meta-search.
google	API key + CX	Highest quality, paid.

Outbound fetches are SSRF-guarded: private/internal IPs are blocked by default (allow_private_hosts: true opts a trusted app out of this), on top of the domain allow/block lists above.

Cross-references

• Module reference: web
• Interactive browser: browser
• Built-in short names: Built-in tools