Search

Hooman ships a web_search tool that normalizes results from a single configured search provider. First-run setup asks you to pick a search provider (DuckDuckGo needs no API key; others are validated with a test search) and writes the search block into config.json. Search is controlled by that block. Each provider page below covers its config options and one example config.

Supported providers

Provider	Runtime id
Brave	brave
DuckDuckGo	duckduckgo
Exa	exa
Firecrawl	firecrawl
LiteLLM	litellm
Serper	serper
Tavily	tavily

Shared shape

{
  "search": {
    "enabled": true,
    "provider": "duckduckgo",
    "duckduckgo": {}
  }
}

• search.enabled (boolean) — turns the web_search tool on or off. Defaults to true for new installs.
• search.provider — the runtime id of one of the providers above. Defaults to duckduckgo.
• A per-provider block carrying that provider’s own options (usually just apiKey; DuckDuckGo needs none).

After setup, change the search provider via /config, hooman config, or the VS Code settings editor — see /config.

Shared search parameters

Regardless of provider, the agent calls web_search with a normalized input shape and gets back normalized results:

Input field	Notes
query	Required search query (max 400 chars).
count	Result count, 1–20 (default 5).
freshness	"day" | "week" | "month" | "year". Mutually exclusive with start_date/end_date.
start_date / end_date	YYYY-MM-DD, must be provided together.
country	Two-letter country code.
safe_search	Boolean.

Every provider returns the same normalized { provider, query, results: [{ title, url, snippet }], metadata } shape; not every provider supports every input field (each provider page notes the gaps).

Search requests use a 20-second timeout by default; firecrawl and litellm get a longer 60-second timeout since a search+scrape round trip can take longer.