deploy(S4): Blender headless pipeline + WebAR client + backend fixes

backend/tools/web_search_tools.py

@@ -11,6 +11,8 @@ import re as _re
 
 # DuckDuckGo's HTML endpoint serves the classic markup; a real browser UA is
 # required or it returns a stripped page with no results.
+_WIKI_UA = "JARVIS_OMEGA/1.0 (+https://huggingface.co/spaces/Jarvis2345/jarvis-cloud)"
+
 _BROWSER_UA = ("Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 "
                "(KHTML, like Gecko) Chrome/120.0 Safari/537.36")
 
@@ -139,6 +141,39 @@ async def search_web(query: str, num_results: int = 5) -> dict:
                 break
         return {"results": results} if results else {"error": "duckduckgo returned no results"}
 
+    def _wikipedia_search():
+        """Free, keyless, genuinely unlimited — and rate-limit free.
+
+        Every commercial option now gates access (Brave/SerpAPI need a card or
+        signup, Gemini search-grounding 429s without billing) and every keyless
+        one throttles (DuckDuckGo challenges automated use, public SearXNG
+        instances 403 the JSON API). Wikipedia's REST/Action API does not, so it
+        is the one source that keeps research working with zero credentials.
+
+        Deliberately placed above the DuckDuckGo scrape: authoritative, fast
+        JSON, and it cannot rate-limit the assistant out of working.
+        """
+        try:
+            api = ("https://en.wikipedia.org/w/api.php?action=query&list=search"
+                   "&srsearch=" + urllib.parse.quote(query) +
+                   f"&srlimit={num_results}&format=json&origin=*")
+            req = urllib.request.Request(api, headers={"User-Agent": _WIKI_UA})
+            with urllib.request.urlopen(req, timeout=12) as resp:
+                data = json.loads(resp.read().decode("utf-8", "replace"))
+            hits = data.get("query", {}).get("search", [])
+            results = []
+            for h in hits[:num_results]:
+                title = h.get("title", "")
+                results.append({
+                    "title": title,
+                    # Wikipedia returns the snippet with HTML search highlighting.
+                    "snippet": _strip_html(h.get("snippet", "")),
+                    "url": "https://en.wikipedia.org/wiki/" + urllib.parse.quote(title.replace(" ", "_")),
+                })
+            return {"results": results} if results else {"error": "wikipedia returned no results"}
+        except Exception as e:
+            return {"error": str(e)}
+
     def _serpapi_search():
         url = ("https://serpapi.com/search.json?q=" + urllib.parse.quote(query)
                + f"&num={num_results}&api_key={urllib.parse.quote(serpapi_key)}")
@@ -183,7 +218,10 @@ async def search_web(query: str, num_results: int = 5) -> dict:
     # rate-limited or misconfigured primary degrades instead of silently
     # returning nothing to the assistant.
     for provider, available in ((_brave_search, brave_api_key),
-                                (_serpapi_search, serpapi_key)):
+                                (_serpapi_search, serpapi_key),
+                                # Keyless and unthrottled — the reason research
+                                # keeps working with no credentials configured.
+                                (_wikipedia_search, True)):
         if not available:
             continue
         out = await asyncio.to_thread(provider)