Web Search Tool

The web_search tool performs web searches and returns ranked results (title, URL, snippet) inline. It uses DuckDuckGo's HTML endpoint keylessly — no API key required.

Usage

The tool accepts a query string and optional max_results:

{
  "query": "Rust async performance tips"
}

Optional parameters:

Field	Type	Default	Description
query	string	—	Search query (also accepts pattern as alias)
max_results	number	config default	Cap returned results (max 20)

Output

{
  "query": "Rust async performance tips",
  "provider": "duckduckgo",
  "count": 5,
  "cached": false,
  "results": [
    { "title": "...", "url": "https://...", "snippet": "..." }
  ]
}

Configuration

Configure via vtcode.toml under [tools.web_search]:

[tools.web_search]
# Provider (currently only "duckduckgo" is supported)
provider = "duckduckgo"

# Default results per call (hard cap: 20)
max_results = 5

# Request timeout in seconds (max: 60)
timeout_secs = 15

# Minimum gap between requests in milliseconds (default: 3000)
cooldown_ms = 3000

# How long results are cached in seconds (default: 300)
cache_ttl_secs = 300

# Session-wide request cap (default: 12)
session_max_requests = 12

Guard Rails

• Cooldown — prevents hammering the endpoint (default 3s between requests)
• Result cache — identical queries served from memory (default 5min TTL, no network call)
• Session cap — limits total outbound requests per session (default 12)
• Timeout — per-request timeout (default 15s, max 60s)
• Results cap — max 20 results per call

Related Tools

• web_fetch — fetch full page content from a specific URL
• defuddle_fetch — fetch and extract structured content from documents