Web Search

Web Search Skill

When to Use

Use this skill when the user wants information that requires looking at the live web.

Trigger Conditions

Use web_search (fast, ~3-5s) when:

• Single-source fact lookup: news, exchange rates, sports scores, weather, stock prices
• One specific product at one named retailer: "what does Bunnings charge for X"
• Simple factual question answerable from one good search result

Use deep_web_research (~60s) when:

• ANY mention of location / "near me" / city name with a discovery intent
• Comparing prices across multiple stores
• Finding local businesses: shops, restaurants, services, pharmacies, tradespeople
• Events: "what's on this weekend", concerts, festivals, markets near a location
• Opening hours of a specific business or category of business
• Jobs: "jobs for nurses in [city]"
• Accommodation: hotels, motels, Airbnb in a location
• Transport timetables, flight searches
• Reviews: "best rated X in [location]"
• Any site that requires a postcode/suburb to show local prices (BWS, Liquorland, Dan Murphy's, Bottlemart)
• Any query where one search result won't cover all local options

Decision Tree

Is there a location / "near me" in the query?
  YES → deep_web_research (it resolves location and finds all local sources)
  NO → Is the user comparing or discovering multiple options?
         YES → deep_web_research
         NO → web_search

What to Say Before deep_web_research

Always tell the user what you're doing before calling the tool, so the ~60s wait feels active:

• Prices: "Looking up prices across local stores in [location]…"
• Events: "Checking what's on in [location] this weekend…"
• Restaurants/food: "Finding [type] options near [location]…"
• Services (plumber, dentist, etc.): "Finding [service] providers near [location]…"
• Hotels/accommodation: "Checking accommodation options in [location]…"
• Transport: "Checking transport options…"
• Generic: "Researching that in [location] — checking multiple sources, this takes about 60 seconds…"

Query Construction for deep_web_research

Include the full location context in the query string:

• Extract city, suburb, postcode from the user message
• If user said "near me" — use their home location from memory
• Good: "cheapest Emu Export beer Geraldton WA 6530"
• Good: "Italian restaurants open now Fitzroy Melbourne 3065"
• Good: "plumbers near Ballarat VIC reviews"
• Bad: "cheapest beer" (no location, no product detail)

What the Framework Does Automatically

When you call deep_web_research, the framework:

1. Classifies query depth (1-5 pages to visit)
2. Resolves location from query text or user's MemPalace home address
3. Runs a DuckDuckGo search for the primary query
4. Queries Google Maps to find ALL local businesses in the category
5. Visits all found pages concurrently with a real stealth browser
6. Automatically fills postcode/suburb gates on retail sites
7. Uses site-internal search if a homepage is detected
8. Returns extracted text for you to synthesise into a ranked answer

You do NOT need to call multiple tools. One deep_web_research call does all of this.

Escalate to OpenClaw instead when

• Task requires login or authenticated session
• Multi-hour background workflow
• Code generation or Home Assistant automation setup
• The task is not web research at all (file editing, system commands, etc.)

Example Calls

{"tool": "deep_web_research", "query": "cheapest Emu Export beer Geraldton WA 6530"}
{"tool": "deep_web_research", "query": "Italian restaurants open now near Fitzroy Melbourne"}
{"tool": "deep_web_research", "query": "plumbers near Geraldton WA reviews phone number"}
{"tool": "deep_web_research", "query": "what's on this weekend Fremantle WA events"}
{"tool": "deep_web_research", "query": "hotels in Broome WA under 150 dollars"}
{"tool": "deep_web_research", "query": "jobs for nurses in Perth WA 2026"}
{"tool": "web_search", "query": "Bitcoin price AUD today"}
{"tool": "web_search", "query": "latest AI news May 2026"}