Upload 2 files

src/api/agent.py

@@ -114,8 +114,7 @@ class AgentResponse(BaseModel):
     primo_search_url: Optional[str] = None
     primo_ai_url: Optional[str] = None
     pubmed_search_url: Optional[str] = None
-    primo_ai_url: Optional[str] = None
-    pubmed_search_url: Optional[str] = None
+    contact_html: Optional[str] = None
 
 
 # ── Constants ─────────────────────────────────────────────────────────────────
@@ -204,6 +203,10 @@ CURRENT_INFO_RE = re.compile(
     r"|news|update|updated|prime minister|ceo|minister|king|ruler|who won)\b",
     re.IGNORECASE,
 )
+DATABASE_RECOMMENDATION_RE = re.compile(
+    r"\b(best database|best databases|which database|which databases|recommended databases?|database should i use|databases should i use|database for|databases for)\b",
+    re.IGNORECASE,
+)
 SUMMARY_RE = re.compile(
     r"\b(summar(y|ize|ise|izing|ising)|overview|explain|what is|what are|describe"
     r"|tell me about|give me a summary|give me an overview|give me a brief"
@@ -728,24 +731,27 @@ def _parse_year_filters(question: str) -> Tuple[Optional[str], Optional[str]]:
 
 
 def _make_primo_boolean_query(context: SearchContextPayload) -> str:
-    """Build synonym-aware PRIMO boolean using shared planner logic."""
+    """
+    FIX: Build a clean boolean for display purposes ONLY.
+    Contains ONLY topic keywords — NO resource-type words.
+    Resource type is expressed as a PRIMO facet in search_primo(), never here.
+    """
     topic = _title_case_topic(context.display_topic or context.topic)
+    # Strip any remaining noise one final time
     topic = _strip_resource_noise(topic)
     if not topic:
         topic = context.topic or "library search"
-    return build_boolean_query(topic)
+    return f'"{_normalize_whitespace(topic)}"'
 
 
 def _primo_clean_url(context: SearchContextPayload) -> str:
-    boolean_query = context.primo_boolean_query or _make_primo_boolean_query(context)
-    return build_primo_discovery_url(
-        boolean_query,
-        context.resource_type,
-        context.peer_reviewed,
-        context.open_access,
-        context.year_from,
-        context.year_to,
-    )
+    """
+    Build the PRIMO discovery URL using the clean topic — matching exactly what the API searches.
+    """
+    clean_topic = _strip_resource_noise(context.topic)
+    if not clean_topic:
+        clean_topic = context.topic
+    return f"{PRIMO_BASE_URL}&query=any,contains,{quote(clean_topic)}&lang=en&mode=advanced&offset=0"
 
 
 def _topic_keywords(text: str) -> List[str]:
@@ -964,10 +970,7 @@ def _extractive_snapshot_from_papers(topic: str, papers: List[dict]) -> str:
 
 # ── PRIMO / PubMed Search Wrappers ────────────────────────────────────────────
 
-from src.api.search import (
-    SearchRequest, search_primo, search_pubmed,
-    build_boolean_query, build_primo_discovery_url, build_pubmed_url,
-)
+from src.api.search import SearchRequest, search_primo, search_pubmed
 
 
 async def _search_primo_catalog(context: SearchContextPayload, limit: int = 5) -> Tuple[dict, str]:
@@ -990,8 +993,7 @@ async def _search_primo_catalog(context: SearchContextPayload, limit: int = 5) -
             open_access=context.open_access,
             year_from=context.year_from,
             year_to=context.year_to,
-            resource_type=context.resource_type,
-            boolean_query=context.primo_boolean_query,
+            resource_type=context.resource_type,  # FIX: was missing — resource type now goes as facet
         ),
         settings.primo_api_key,
     )
@@ -1017,7 +1019,9 @@ async def _search_pubmed_articles(context: SearchContextPayload, limit: int = 5)
             year_to=context.year_to,       # FIX: now actually used in search.py
         )
     )
-    pubmed_url = build_pubmed_url(clean_topic, context.year_from, context.year_to)
+    pubmed_url = f"https://pubmed.ncbi.nlm.nih.gov/?term={quote(clean_topic)}"
+    if context.year_from and context.year_to:
+        pubmed_url += f"&filter=datesearch.y_custom&dates.from={context.year_from}&dates.to={context.year_to}"
     return result, pubmed_url
 
 
@@ -1026,102 +1030,134 @@ async def _search_pubmed_articles(context: SearchContextPayload, limit: int = 5)
 def _tool_urls(context: SearchContextPayload) -> Dict[str, str]:
     ai_query = quote(context.ai_tool_query or context.display_topic or context.topic)
     clean_topic = _strip_resource_noise(context.topic)
-    semantic_q = quote((context.ai_tool_query or context.display_topic or context.topic).replace(" ", "+"))
     return {
         "leapspace": f"https://www.sciencedirect.com/leapspace?query={ai_query}",
         "scopus_ai": f"https://www-scopus-com.khalifa.idm.oclc.org/pages/ai?query={ai_query}",
-        "ebsco_ai": f"https://research-ebsco-com.khalifa.idm.oclc.org/c/ntdlkp/search?q={ai_query}",
-        "primo_ai": f"{PRIMO_AI_URL}{ai_query}",
         "consensus": f"https://consensus.app/results/?q={ai_query}",
-        "openalex": f"https://openalex.org/works?page=1&sort=relevance_score:desc&search.semantic={semantic_q}",
-        "scispace": f"https://scispace.com/search?q={ai_query}",
-        "perplexity": f"https://www.perplexity.ai/search?q={ai_query}",
         "primo_discovery": _primo_clean_url(context),
-        "pubmed": build_pubmed_url(clean_topic, context.year_from, context.year_to) if context.intent == "search_medical" else "",
     }
 
 
 def _database_guidance_for_topic(context: SearchContextPayload) -> str:
     topic = _escape(context.display_topic or context.topic)
-    urls = _tool_urls(context)
-    ai_links = (
-        f'<a href="{urls["leapspace"]}" target="_blank" style="color:#9a3412;font-weight:700;text-decoration:none">LeapSpace</a> · '
-        f'<a href="{urls["scopus_ai"]}" target="_blank" style="color:#1e40af;font-weight:700;text-decoration:none">Scopus AI</a> · '
-        f'<a href="{urls["ebsco_ai"]}" target="_blank" style="color:#9d174d;font-weight:700;text-decoration:none">EBSCO AI</a> · '
-        f'<a href="{urls["primo_ai"]}" target="_blank" style="color:#5b21b6;font-weight:700;text-decoration:none">PRIMO AI Assistant</a> · '
-        f'<a href="{urls["consensus"]}" target="_blank" style="color:#86198f;font-weight:700;text-decoration:none">Consensus</a> · '
-        f'<a href="{urls["openalex"]}" target="_blank" style="color:#0c4a6e;font-weight:700;text-decoration:none">OpenAlex</a>'
-    )
-    more_results = f'<a href="{urls["primo_discovery"]}" target="_blank" style="color:#5b21b6;font-weight:700;text-decoration:none">PRIMO Library Discovery</a>'
-    if context.intent == "search_medical" and urls["pubmed"]:
-        more_results += f' · <a href="{urls["pubmed"]}" target="_blank" style="color:#065f46;font-weight:700;text-decoration:none">PubMed</a>'
+    if context.intent == "search_medical":
+        return (
+            f"For deeper research on <strong>{topic}</strong>, the KU-subscribed AI research tools and freemium tools are listed at the end of this answer. "
+            "Direct links to PRIMO Library Discovery and PubMed are also provided there. "
+            "Some freemium tools may require account creation."
+        )
     return (
-        f'<strong>AI tools:</strong> {ai_links}<br>'
-        f'<strong>More results:</strong> {more_results}'
+        f"For deeper research on <strong>{topic}</strong>, the KU-subscribed AI research tools and freemium tools are listed at the end of this answer. "
+        "Direct links to PRIMO Library Discovery and, for medical topics, PubMed are also provided there. "
+        "Some freemium tools may require account creation."
     )
 
 
+def _is_database_recommendation_question(question: str) -> bool:
+    return bool(DATABASE_RECOMMENDATION_RE.search(question or ""))
 
 
-GENERAL_CONTACT_HTML = (
-    'General queries: <a href="https://library.ku.ac.ae/AskUs" target="_blank">Ask a Librarian</a> '
-    '· <a href="mailto:libse@ku.ac.ae">libse@ku.ac.ae</a>'
-)
+def _primary_contact_for_question(question: str, intent: str) -> dict:
+    q = (question or '').lower()
+    walter = {
+        'label': 'Systems & website support',
+        'name': 'Walter Brian Hall',
+        'role': 'Digital & Technology Services Librarian / Systems Librarian',
+        'email': 'walter.hall@ku.ac.ae',
+        'phone': '+971 2 312 3163',
+    }
+    rani = {
+        'label': 'E-resources support',
+        'name': 'Rani Anand',
+        'role': 'E-Resources Librarian',
+        'email': 'rani.anand@ku.ac.ae',
+        'phone': '+971 2 312 3935',
+    }
+    nikesh = {
+        'label': 'Research support',
+        'name': 'Nikesh Narayanan',
+        'role': 'Research & Access Services Librarian',
+        'email': 'nikesh.narayanan@ku.ac.ae',
+        'phone': '+971 2 312 3980',
+    }
+    muna = {
+        'label': 'Public services',
+        'name': 'Muna Ahmad Mohammad Al Blooshi',
+        'role': 'Manager, Public Services',
+        'email': 'muna.alblooshi@ku.ac.ae',
+        'phone': '+971 2 312 4212',
+    }
 
+    if any(t in q for t in ['system', 'systems', 'website', 'web site', 'technical', 'technology', 'digital', 'error', 'bug', 'broken page', 'page not loading']):
+        return walter
+    if any(t in q for t in ['database access', 'off campus', 'remote access', 'proxy', 'login problem', 'login issue', 'cannot access', "can't access", 'full text', 'libkey', 'vendor', 'e-resource', 'eresource', 'subscription access']):
+        return rani
+    if intent in {'search_academic', 'search_medical'} or any(t in q for t in ['research', 'literature', 'systematic review', 'orcid', 'open access', 'apc', 'bibliometric', 'scopus', 'scival', 'impact', 'citation', 'ai tool', 'best database', 'best databases', 'which database', 'which databases']):
+        return nikesh
+    return muna
 
-def _contact_footer(question: str, intent: str, context: Optional[SearchContextPayload] = None) -> str:
-    q = (question or "").lower()
-    lines = [GENERAL_CONTACT_HTML]
-    # Public services for general library help
-    if intent == "library_info":
-        lines.append('Public services: Muna Al Blooshi · <a href="mailto:muna.alblooshi@ku.ac.ae">muna.alblooshi@ku.ac.ae</a> · +971 2 312 4212')
-    # E-resources contact for access/search/database/full-text issues
-    if intent in {"search_academic", "search_medical"} or any(t in q for t in ["database", "e-resource", "eresource", "access", "full text", "proxy", "login", "remote access", "primo", "pubmed"]):
-        lines.append('E-resources: Rani Anand · <a href="mailto:rani.anand@ku.ac.ae">rani.anand@ku.ac.ae</a> · +971 2 312 3935')
-    # Research support for scholarly/research questions
-    if intent in {"search_academic", "search_medical"} or any(t in q for t in ["research", "literature review", "citation", "orcid", "open access", "refworks", "bibliometrics"]):
-        lines.append('Research support: Nikesh Narayanan · <a href="mailto:nikesh.narayanan@ku.ac.ae">nikesh.narayanan@ku.ac.ae</a> · +971 2 312 3980')
-    # Systems / technology contact
-    if any(t in q for t in ["system", "systems", "website", "technology", "technical", "error", "broken"]):
-        lines.append('Systems: Walter Brian Hall · <a href="mailto:walter.hall@ku.ac.ae">walter.hall@ku.ac.ae</a> · +971 2 312 3163')
-    dedup = []
-    seen = set()
-    for line in lines:
-        if line not in seen:
-            seen.add(line)
-            dedup.append(line)
-    html_lines = ''.join(f'<div>{line}</div>' for line in dedup)
+
+def _contact_footer_html(question: str, intent: str) -> str:
+    primary = _primary_contact_for_question(question, intent)
     return (
-        '<div style="margin-top:12px;padding:10px 12px;background:#f8fafc;border:1px solid #dbeafe;border-radius:10px;font-size:.8rem;color:#1e3a8a">'
-        '<strong>Library contact points</strong>' + html_lines + '</div>'
+        '<div style="margin-top:14px;padding:12px 14px;background:#f8fafc;border:1px solid #dbeafe;border-radius:12px;font-size:.8rem;line-height:1.6">'
+        '<strong>Library contact points</strong><br>'
+        f'General help: <a href="{ASK_LIBRARIAN_URL}" target="_blank" style="color:#003366;font-weight:700">Ask a Librarian</a> · '
+        '<a href="mailto:libse@ku.ac.ae" style="color:#003366">libse@ku.ac.ae</a><br>'
+        f'{_escape(primary["label"])}: <strong>{_escape(primary["name"])}</strong> — {_escape(primary["role"])} · '
+        f'<a href="mailto:{primary["email"]}" style="color:#003366">{_escape(primary["email"])}</a> · '
+        f'<a href="tel:{primary["phone"].replace(" ", "")}" style="color:#003366">{_escape(primary["phone"])}</a>'
+        '</div>'
     )
 
 
-def _scholarly_trace_html(question: str, context: SearchContextPayload, mode: str) -> str:
-    topic = _escape(context.display_topic or context.topic)
-    ai_query = _escape(context.ai_tool_query or question)
-    primo_boolean = _escape(context.primo_boolean_query or _make_primo_boolean_query(context))
-    retrieval = (
-        'I corrected and normalized your question, extracted the main topic, expanded close synonyms with OR, connected concept groups with AND, '
-        'then searched Semantic Scholar for scholarly evidence and PRIMO for KU-accessible results.'
-        if mode == 'snapshot' else
-        'I corrected and normalized your question, extracted the main topic, expanded close synonyms with OR, connected concept groups with AND, '
-        f'then searched {"PubMed" if context.source == "pubmed" else "PRIMO Library Discovery"} using cleaned topic terms plus the active filters.'
-    )
-    extra = ''
-    if context.intent == 'search_medical':
-        extra = f'<div><strong>PubMed search link:</strong> <a href="{build_pubmed_url(context.topic, context.year_from, context.year_to)}" target="_blank">Open PubMed for this topic</a></div>'
-    return (
-        '<div style="margin:0 0 10px 0;padding:10px 12px;background:#f8fafc;border:1px solid #d1d5db;border-radius:10px;font-size:.8rem;color:#334155">'
-        '<div style="font-weight:700;color:#003366;margin-bottom:6px">Search strategy used</div>'
-        f'<div><strong>Your question:</strong> {_escape(question)}</div>'
-        f'<div><strong>Interpreted topic:</strong> {topic}</div>'
-        f'<div><strong>PRIMO Library Discovery query:</strong> <code>{primo_boolean}</code></div>'
-        f'<div><strong>AI tools query:</strong> {_escape(ai_query)}</div>'
-        f'{extra}'
-        f'<div><strong>How I found this answer:</strong> {retrieval}</div>'
-        '</div>'
+async def _database_recommendation_answer(question: str, history: List[ChatMessage], model: str) -> Tuple[str, List[dict]]:
+    try:
+        from app import rag_service
+        rag_query = f"{question} recommended databases by subject Khalifa University Library databases"
+        rag_results = await rag_service.hybrid_search(rag_query, top_k=5)
+    except Exception as e:
+        logger.error(f"Database recommendation search failed: {e}")
+        rag_results = []
+
+    if not rag_results:
+        answer = (
+            'I could not pull the subject database guidance from the library knowledge base right now.<br><br>'
+            'A good general starting point is <strong>PRIMO Library Discovery</strong>, then move to subject databases such as Scopus, Web of Science, IEEE Xplore, ACM Digital Library, PubMed, or Embase depending on your topic.'
+        )
+        return answer + _source_badge('ku_library', 'https://library.ku.ac.ae/eresources'), []
+
+    context_text = "\n\n---\n\n".join(r['content'] for r in rag_results)
+    settings = get_settings()
+    if not settings.openai_api_key:
+        answer = 'I found KU Library database guidance for this subject in the knowledge base.'
+        answer += _source_badge('rag', 'https://library.ku.ac.ae/eresources')
+        sources = [{'title': r.get('title', 'Library guide'), 'url': r.get('source', '')} for r in rag_results if r.get('source')]
+        return answer, sources
+
+    system = (
+        'You are LibBee, the Khalifa University Library AI Assistant. '
+        'Answer as an expert academic librarian using ONLY the context provided. '
+        "Recommend the best KU databases for the user's subject. "
+        'Start with one short sentence. Then list 3 to 6 databases using HTML <br>, with each database name in <strong> and a brief explanation of why or when to use it. '
+        'Where appropriate, include PRIMO Library Discovery as a starting point and then subject-specific databases. '
+        'End with one short sentence offering to search directly in PRIMO Library Discovery.'
     )
+    try:
+        llm = _get_llm(model, temperature=0.2, max_tokens=420)
+        response = await llm.ainvoke([
+            {'role': 'system', 'content': system + "\n\nCONTEXT:\n" + context_text},
+            {'role': 'user', 'content': question},
+        ])
+        answer = response.content.strip()
+    except Exception as e:
+        logger.error(f"Database recommendation answer failed: {e}")
+        answer = 'I found subject-specific database guidance in the library knowledge base, but I had trouble formatting it into a recommendation right now.'
+
+    answer += _source_badge('rag', 'https://library.ku.ac.ae/eresources')
+    sources = [{'title': r.get('title', 'Library guide'), 'url': r.get('source', '')} for r in rag_results if r.get('source')]
+    return answer, sources
+
 
 # ── Research Snapshot ─────────────────────────────────────────────────────────
 
@@ -1162,7 +1198,6 @@ async def _research_snapshot(context: SearchContextPayload, model: str) -> Tuple
         answer += _source_badge("semantic", "https://www.semanticscholar.org/")
         if search_results:
             answer += _source_badge("primo", primo_url)
-        answer = _scholarly_trace_html(topic, context, "snapshot") + answer + "<br><br>" + _database_guidance_for_topic(context) + _contact_footer(topic, context.intent, context)
         references = [{"title": p["title"], "url": p["link"]} for p in papers[:6]]
         return answer, search_results, references
 
@@ -1178,7 +1213,6 @@ async def _research_snapshot(context: SearchContextPayload, model: str) -> Tuple
             f"<strong>AI sources:</strong> {_database_guidance_for_topic(context)}"
         )
     answer += _source_badge("primo", primo_url)
-    answer = _scholarly_trace_html(topic, context, "snapshot") + answer + "<br><br>" + _contact_footer(topic, context.intent, context)
     return answer, search_results, []
 
 
@@ -1269,8 +1303,6 @@ def _detect_follow_up(question: str, client_state: Optional[ClientStatePayload])
         return False, None, None
     explicit_action = (client_state.follow_up_action if client_state else None) or None
     if explicit_action:
-        if explicit_action == "new_search":
-            return False, None, None
         return True, explicit_action, base_context
     action = _parse_refinement_action(question)
     if not action:
@@ -1315,24 +1347,28 @@ def _apply_follow_up_action(base_context: SearchContextPayload, action: Optional
 
 async def _generate_topic_follow_ups(topic: str, model: str) -> List[dict]:
     """
-    Generate 3 topic-specific follow-up questions that ALWAYS preserve the original topic anchor.
-    These should start a fresh contextual search, not a refinement action.
+    FIX: Generate 3 topic-specific intellectual follow-up questions using LLM.
+    These are ASPECT-based (related subtopics, methodological angles, applications)
+    not generic filter suggestions like "peer reviewed only".
     """
     settings = get_settings()
     if not settings.openai_api_key:
         return []
     try:
-        llm = _get_llm(model, temperature=0.5, max_tokens=220)
+        llm = _get_llm(model, temperature=0.5, max_tokens=200)
         response = await llm.ainvoke([
             {"role": "system", "content": (
                 "You are a research librarian helping a user explore a topic more deeply. "
-                "Given a research topic, generate exactly 3 short follow-up questions that explore different aspects of the SAME topic. "
-                "Every suggestion must keep the original topic explicitly present. "
-                "For example, if the topic is 'global warming', acceptable suggestions include "
-                "'What are the main adaptation strategies for global warming?' and "
-                "'How does global warming affect water security?'. "
-                "Do NOT suggest generic filters like peer reviewed or last 5 years. "
-                "Return ONLY a JSON array of 3 strings."
+                "Given a research topic, generate exactly 3 short follow-up questions that explore "
+                "DIFFERENT ASPECTS of the topic — such as subtopics, methodological angles, "
+                "applications, comparisons, or related fields. "
+                "Do NOT suggest filter actions like peer-reviewed, date ranges, or format changes. "
+                "Return ONLY a JSON array of 3 strings, each under 12 words. "
+                "No explanation, no preamble, no markdown.\n"
+                "Example for 'quantum computing':\n"
+                '[\"What are the main hardware approaches in quantum computing?\", '
+                '"How does quantum error correction work?\", '
+                '"Applications of quantum computing in cryptography\"]'
             )},
             {"role": "user", "content": f"Topic: {topic}"},
         ])
@@ -1342,26 +1378,16 @@ async def _generate_topic_follow_ups(topic: str, model: str) -> List[dict]:
         questions = json.loads(raw)
         if not isinstance(questions, list):
             return []
-        out = []
-        topic_l = (topic or "").lower()
-        for q in questions[:3]:
-            if not isinstance(q, str) or not q.strip():
-                continue
-            clean = q.strip().rstrip('?')
-            if topic_l not in clean.lower():
-                clean = f"{clean} for {topic}"
-            full_q = clean + "?"
-            out.append({
-                "label": full_q,
-                "question": f"Find research articles on {full_q}",
-            })
-        return out
+        return [
+            {"label": q.strip("?") + "?", "question": f"Find research articles on {_normalize_whitespace(q.strip().rstrip("?"))} in the context of {topic}", "action": None}
+            for q in questions[:3]
+            if isinstance(q, str) and q.strip()
+        ]
     except Exception as e:
         logger.warning(f"Topic follow-up generation failed: {e}")
         return []
 
 
-
 async def _search_follow_up(
     context: SearchContextPayload, model: str, summary_mode: bool = False
 ) -> Tuple[str, List[dict]]:
@@ -1513,7 +1539,7 @@ async def _run_search_mode(
     search_results = result.get("results", []) or []
     total = result.get("total", len(search_results))
 
-    answer = _scholarly_trace_html(question, context, "search") + _search_answer_intro(context, is_follow_up)
+    answer = _search_answer_intro(context, is_follow_up)
     if filters:
         answer += f"<br><br><strong>Current filters:</strong> {_escape(filters)}."
     if search_results:
@@ -1526,9 +1552,7 @@ async def _run_search_mode(
     if context.intent == "search_medical" and context.source != "pubmed":
         answer += "<br><br>For the strongest medical evidence trail, I also recommend checking <strong>PubMed</strong> or <strong>Embase</strong>."
 
-    answer += "<br><br>" + _database_guidance_for_topic(context)
     answer += _source_badge(source_name, source_url)
-    answer += _contact_footer(question, context.intent, context)
     sources = []
     for item in search_results[:6]:
         link = item.get("primo_url") or item.get("link")
@@ -1595,6 +1619,7 @@ async def agent_query(request: Request, req: AgentRequest):
             response_time=time.time() - start_time,
             follow_up_question=follow_up_question,
             follow_up_suggestions=follow_up_suggestions,
+            contact_html=_contact_footer_html(question, "library_info"),
         )
 
     staff_role_match = match_staff_role(question) if should_attempt_staff_lookup(question) else None
@@ -1609,6 +1634,7 @@ async def agent_query(request: Request, req: AgentRequest):
             response_time=time.time() - start_time,
             follow_up_question=follow_up_question,
             follow_up_suggestions=follow_up_suggestions,
+            contact_html=_contact_footer_html(question, "library_info"),
         )
 
     if _is_greeting_menu_followup(question, history):
@@ -1626,7 +1652,7 @@ async def agent_query(request: Request, req: AgentRequest):
     if _looks_library_hours_question(question):
         follow_up_question, follow_up_suggestions = _library_follow_up(question)
         return AgentResponse(
-            answer=_hours_answer() + _source_badge("ku_library", LIBRARY_HOURS_URL) + _contact_footer(question, "library_info"),
+            answer=_hours_answer() + _source_badge("ku_library", LIBRARY_HOURS_URL),
             intent="library_info",
             tools_used=["hours_hardcoded"],
             ask_librarian=False,
@@ -1634,12 +1660,13 @@ async def agent_query(request: Request, req: AgentRequest):
             response_time=time.time() - start_time,
             follow_up_question=follow_up_question,
             follow_up_suggestions=follow_up_suggestions,
+            contact_html=_contact_footer_html(question, "library_info"),
         )
 
     if _looks_campus_question(question):
         follow_up_question, follow_up_suggestions = _library_follow_up(question)
         return AgentResponse(
-            answer=_campus_answer() + _source_badge("ku_library", "https://library.ku.ac.ae/") + _contact_footer(question, "library_info"),
+            answer=_campus_answer() + _source_badge("ku_library", "https://library.ku.ac.ae/"),
             intent="library_info",
             tools_used=["campus_hardcoded"],
             ask_librarian=False,
@@ -1647,6 +1674,27 @@ async def agent_query(request: Request, req: AgentRequest):
             response_time=time.time() - start_time,
             follow_up_question=follow_up_question,
             follow_up_suggestions=follow_up_suggestions,
+            contact_html=_contact_footer_html(question, "library_info"),
+        )
+
+    if _is_database_recommendation_question(question):
+        answer, sources = await _database_recommendation_answer(question, history, model)
+        follow_up_question = "Would you like me to search one of these databases directly or show the AI research tools for this topic?"
+        follow_up_suggestions = [
+            {"label": "Search PRIMO Library Discovery", "question": "Search PRIMO Library Discovery for this topic"},
+            {"label": "Show AI research tools", "question": "Show the AI research tools for this topic"},
+        ]
+        return AgentResponse(
+            answer=answer,
+            intent="library_info",
+            tools_used=["database_recommendation_rag"],
+            sources=sources,
+            ask_librarian=False,
+            model_used=model,
+            response_time=time.time() - start_time,
+            follow_up_question=follow_up_question,
+            follow_up_suggestions=follow_up_suggestions,
+            contact_html=_contact_footer_html(question, "library_info"),
         )
 
     is_follow_up, follow_up_action, base_context = _detect_follow_up(question, client_state)
@@ -1683,12 +1731,13 @@ async def agent_query(request: Request, req: AgentRequest):
                 ai_tool_query=context.ai_tool_query,
                 primo_boolean_query=context.primo_boolean_query,
                 primo_search_url=_primo_clean_url(context),
-                primo_ai_url=_tool_urls(context).get("primo_ai"),
-                pubmed_search_url=_tool_urls(context).get("pubmed"),
                 follow_up_question=follow_up_question,
                 follow_up_suggestions=follow_up_suggestions,
                 search_context=_context_to_dict(context),
                 is_follow_up=True,
+                primo_ai_url=f"{PRIMO_AI_URL}{quote(context.ai_tool_query or context.topic)}",
+                pubmed_search_url=(f"https://pubmed.ncbi.nlm.nih.gov/?term={quote(context.topic)}" if context.intent == "search_medical" else None),
+                contact_html=_contact_footer_html(question, context.intent),
             )
 
         answer, search_results, sources, source_url = await _run_search_mode(question, context, model, True)
@@ -1707,12 +1756,13 @@ async def agent_query(request: Request, req: AgentRequest):
             ai_tool_query=context.ai_tool_query,
             primo_boolean_query=context.primo_boolean_query,
             primo_search_url=source_url,
-            primo_ai_url=_tool_urls(context).get("primo_ai"),
-            pubmed_search_url=_tool_urls(context).get("pubmed"),
             follow_up_question=follow_up_question,
             follow_up_suggestions=follow_up_suggestions,
             search_context=_context_to_dict(context),
             is_follow_up=True,
+            primo_ai_url=f"{PRIMO_AI_URL}{quote(context.ai_tool_query or context.topic)}",
+            pubmed_search_url=(f"https://pubmed.ncbi.nlm.nih.gov/?term={quote(context.topic)}" if context.intent == "search_medical" else None),
+            contact_html=_contact_footer_html(question, context.intent),
         )
 
     cls_result = await _llm_classify(question, history, model)
@@ -1749,13 +1799,11 @@ async def agent_query(request: Request, req: AgentRequest):
                 f'🔗 <a href="{ASK_LIBRARIAN_URL}" target="_blank">Ask a Librarian</a>'
             )
             answer += _source_badge("ku_library", ASK_LIBRARIAN_URL)
-            answer += _contact_footer(question, "library_info")
             sources = []
             tools_used = ["rag_no_results"]
         else:
             answer = await _rag_answer(question, rag_results, history, model)
             answer += _source_badge("rag", "https://library.ku.ac.ae/")
-            answer += _contact_footer(question, "library_info")
             sources = [{"title": r.get("title", "Library KB"), "url": r.get("source", "")} for r in rag_results if r.get("source")]
             tools_used = ["rag_hybrid_search"]
 
@@ -1770,6 +1818,7 @@ async def agent_query(request: Request, req: AgentRequest):
             response_time=time.time() - start_time,
             follow_up_question=follow_up_question,
             follow_up_suggestions=follow_up_suggestions,
+            contact_html=_contact_footer_html(question, "library_info"),
         )
 
     if intent in ("search_academic", "search_medical"):
@@ -1798,6 +1847,8 @@ async def agent_query(request: Request, req: AgentRequest):
 
         if summary_mode or deep_research_mode:
             answer, search_results, sources = await _research_snapshot(context, model)
+            if deep_research_mode and "Research guide" not in answer:
+                answer += f"<br><br><strong>Research guide:</strong> {_database_guidance_for_topic(context)}"
             follow_up_question, follow_up_suggestions = await _search_follow_up(context, model, summary_mode=True)
             _safe_metrics_increment("search_handoffs")
             return AgentResponse(
@@ -1814,11 +1865,12 @@ async def agent_query(request: Request, req: AgentRequest):
                 ai_tool_query=context.ai_tool_query,
                 primo_boolean_query=context.primo_boolean_query,
                 primo_search_url=_primo_clean_url(context),
-                primo_ai_url=_tool_urls(context).get("primo_ai"),
-                pubmed_search_url=_tool_urls(context).get("pubmed"),
                 follow_up_question=follow_up_question,
                 follow_up_suggestions=follow_up_suggestions,
                 search_context=_context_to_dict(context),
+                primo_ai_url=f"{PRIMO_AI_URL}{quote(context.ai_tool_query or context.topic)}",
+                pubmed_search_url=(f"https://pubmed.ncbi.nlm.nih.gov/?term={quote(context.topic)}" if intent == "search_medical" else None),
+                contact_html=_contact_footer_html(question, intent),
             )
 
         answer, search_results, sources, source_url = await _run_search_mode(question, context, model, False)
@@ -1837,11 +1889,12 @@ async def agent_query(request: Request, req: AgentRequest):
             ai_tool_query=context.ai_tool_query,
             primo_boolean_query=context.primo_boolean_query,
             primo_search_url=source_url,
-            primo_ai_url=_tool_urls(context).get("primo_ai"),
-            pubmed_search_url=_tool_urls(context).get("pubmed"),
             follow_up_question=follow_up_question,
             follow_up_suggestions=follow_up_suggestions,
             search_context=_context_to_dict(context),
+            primo_ai_url=f"{PRIMO_AI_URL}{quote(context.ai_tool_query or context.topic)}",
+            pubmed_search_url=(f"https://pubmed.ncbi.nlm.nih.gov/?term={quote(context.topic)}" if intent == "search_medical" else None),
+            contact_html=_contact_footer_html(question, intent),
         )
 
     if intent == "general_recent":
@@ -1855,6 +1908,7 @@ async def agent_query(request: Request, req: AgentRequest):
             response_time=time.time() - start_time,
             follow_up_question=follow_up_question,
             follow_up_suggestions=follow_up_suggestions,
+            contact_html=_contact_footer_html(question, "general_recent"),
         )
 
     answer = await _web_search_answer(question, history, model) + _source_badge("web")
@@ -1867,6 +1921,7 @@ async def agent_query(request: Request, req: AgentRequest):
         response_time=time.time() - start_time,
         follow_up_question=follow_up_question,
         follow_up_suggestions=follow_up_suggestions,
+        contact_html=_contact_footer_html(question, "general"),
     )