Web Search and Fetch

Triggerfish gives your agent access to the internet through two tools: web_search for finding information and web_fetch for reading web pages. Together they let the agent research topics, look up documentation, check current events, and pull in data from the web -- all under the same policy enforcement as every other tool.

Tools

web_search

Search the web. Returns titles, URLs, and snippets.

Parameter	Type	Required	Description
query	string	yes	Search query. Be specific -- include relevant keywords, names, or dates for better results.
max_results	number	no	Maximum results to return (default: 5, max: 20).

Example response:

Search results for "deno sqlite module":

1. @db/sqlite - Deno SQLite bindings
   https://jsr.io/@db/sqlite
   Fast SQLite3 bindings for Deno using FFI...

2. Deno SQLite Guide
   https://docs.deno.com/examples/sqlite
   How to use SQLite with Deno...

web_fetch

Fetch and extract readable content from a URL. Returns article text by default using Mozilla Readability.

Parameter	Type	Required	Description
url	string	yes	The URL to fetch. Use URLs from web_search results.
mode	string	no	Extraction mode: readability (default, article text) or raw (full HTML).

Extraction modes:

• readability (default) -- Extracts the main article content, stripping navigation, ads, and boilerplate. Best for news articles, blog posts, and documentation.
• raw -- Returns the full HTML. Use when readability extraction returns too little content (e.g., single-page apps, dynamic content).

How the Agent Uses Them

The agent follows a search-then-fetch pattern:

1. Use web_search to find relevant URLs
2. Use web_fetch to read the most promising pages
3. Synthesize the information and cite sources

When answering with web information, the agent cites source URLs inline so they are visible across all channels (Telegram, Slack, CLI, etc.).

Configuration

Web search requires a search provider. Configure it in triggerfish.yaml:

web:
  search:
    provider: brave # Search backend (brave is the default)
    api_key: your-api-key # Brave Search API key

Key	Type	Description
web.search.provider	string	Search backend. Currently supported: brave.
web.search.api_key	string	API key for the search provider.

If no search provider is configured, web_search returns an error

message telling the agent that search is unavailable. web_fetch works independently -- it does not require a search provider. :::

Security

• All fetched URLs pass through SSRF prevention: DNS is resolved first and checked against a hardcoded IP denylist. Private/reserved IP ranges are always blocked.
• Fetched content is classified and contributes to session taint like any other tool response.
• The PRE_TOOL_CALL hook fires before each fetch, and POST_TOOL_RESPONSE fires after, so custom policy rules can restrict which domains the agent accesses.