diff --git a/backend/agent/api.py b/backend/agent/api.py
index 3a0a7dc..b1a87bc 100644
--- a/backend/agent/api.py
+++ b/backend/agent/api.py
@@ -631,9 +631,10 @@ def respond_to_confirmation(run_id: str, confirmation_id: str, request: Confirma
 async def route_workflow(request: RouterSelectRequest):
     db = SessionLocal()
     try:
+        ensure_global_knowledge_store()
         return await select_workflow(
             db, message=request.message, recent_messages=request.recent_messages[-4:], attachments=request.attachments,
-            library_slug=request.library_slug, router_model=request.model, chat_model=request.chat_model,
+            library_slug=request.library_slug or global_library_slug(), router_model=request.model, chat_model=request.chat_model,
             web_search_enabled=request.web_search_enabled,
             confidence_threshold=request.confidence_threshold,
         )
diff --git a/backend/agent/workflows/builtins.py b/backend/agent/workflows/builtins.py
index 29c15b0..0e04dfb 100644
--- a/backend/agent/workflows/builtins.py
+++ b/backend/agent/workflows/builtins.py
@@ -277,10 +277,10 @@ BUILTIN_WORKFLOWS = [
     {"slug": "input-output", "name": "Input -> Output", "description": "Normal Heimgeist chat with optional compatibility context.", "routing_description": "Choose only when no external tool is needed: conversational replies, writing, editing, reasoning, brainstorming, explanation, or stable knowledge the model can answer without retrieval. Do not choose when the standalone request needs facts outside the conversation/model context, source grounding, live or recently changed information, local database retrieval, attachment analysis, or knowledge-saving actions.", "routing_examples": ["Explain this concept", "Draft a reply", "Rewrite this paragraph", "What do you mean by that?"], "estimated_cost_class": "low", "required_capabilities": ["chat"], "graph": DIRECT},
     {"slug": "chat-memory-answer", "name": "Chat Memory Answer", "description": "Search previous Heimgeist chats and answer from matching turns.", "routing_description": "Choose when the standalone request asks Heimgeist to use past chats, earlier conversations, remembered discussion context, previous decisions, or topics discussed before. This workflow searches previous chat turns before answering. Do not choose for general factual questions that merely resemble something discussed before; choose Web Answer, Knowledge Answer, or direct chat based on the requested evidence.", "routing_examples": ["What music did we talk about before?", "What did we decide earlier about packaging?", "Use our previous chat about Ollama settings"], "estimated_cost_class": "medium", "required_capabilities": ["chat_memory", "chat"], "graph": CHAT_MEMORY},
     {"slug": "recent-chat-memory-answer", "name": "Recent Chat Memory Answer", "description": "Answer from the most recent previous Heimgeist chat.", "routing_description": "Choose when the standalone request specifically asks about the last, previous, latest, or most recent earlier Heimgeist chat or conversation. This workflow retrieves the most recent previous chat turn instead of doing a relevance search.", "routing_examples": ["What was the last conversation about?", "Worum ging es im letzten Gespräch?", "Continue from the previous chat"], "estimated_cost_class": "medium", "required_capabilities": ["chat_memory", "chat"], "graph": RECENT_CHAT_MEMORY},
-    {"slug": "knowledge-answer", "name": "Knowledge Answer", "description": "Answer from a selected local knowledge database.", "routing_description": "Choose when the standalone request asks about or should be grounded in the selected local knowledge database, and current web evidence is not required. This workflow retrieves local knowledge before answering.", "routing_examples": ["What do my notes say about this?", "Summarize the selected database entry"], "estimated_cost_class": "medium", "required_capabilities": ["rag"], "graph": KNOWLEDGE},
+    {"slug": "knowledge-answer", "name": "Knowledge Answer", "description": "Answer from local knowledge.", "routing_description": "Choose when the standalone request asks about or should be grounded in Heimgeist's local Knowledge store, and current web evidence is not required. This workflow retrieves local knowledge before answering.", "routing_examples": ["What do my notes say about this?", "Summarize the saved knowledge entry"], "estimated_cost_class": "medium", "required_capabilities": ["rag"], "graph": KNOWLEDGE},
     {"slug": "web-answer", "name": "Web Answer", "description": "Search, fetch, rank, and answer from current web sources.", "routing_description": "Choose when the standalone request needs web evidence and can likely be answered by one bounded search/fetch/rank pass: facts outside the conversation/model context, source-grounded lookup, public information that may have changed, or a follow-up that inherits such a web-grounded topic. This is the default web workflow for direct answer requests. Prefer it over Research unless the request itself requires investigation, comparison, validation, or broad synthesis.", "routing_examples": ["Look this up online", "What is the current status?", "Find the latest information about this topic", "What happened with that since then?"], "estimated_cost_class": "medium", "required_capabilities": ["web"], "graph": web_graph()},
     {"slug": "vision-answer", "name": "Vision Answer", "description": "Analyze attachments and answer with a vision model.", "routing_description": "Use when image or document attachments must be analyzed.", "routing_examples": ["What is shown in this image?"], "estimated_cost_class": "medium", "required_capabilities": ["vision"], "graph": VISION},
-    {"slug": "knowledge-web-answer", "name": "Knowledge + Web Answer", "description": "Combine local knowledge and current web evidence.", "routing_description": "Choose when both capabilities are needed: the selected local knowledge database is relevant, and the standalone request also requires web evidence or updated external facts.", "routing_examples": ["Compare my notes with the latest information", "Check my saved source against current web evidence"], "estimated_cost_class": "high", "required_capabilities": ["rag", "web"], "graph": KNOWLEDGE_WEB},
+    {"slug": "knowledge-web-answer", "name": "Knowledge + Web Answer", "description": "Combine local knowledge and current web evidence.", "routing_description": "Choose when both capabilities are needed: Heimgeist's local Knowledge store is relevant, and the standalone request also requires web evidence or updated external facts.", "routing_examples": ["Compare my notes with the latest information", "Check my saved source against current web evidence"], "estimated_cost_class": "high", "required_capabilities": ["rag", "web"], "graph": KNOWLEDGE_WEB},
     {"slug": "remember-this", "name": "Remember This", "description": "Save a selected chat message to knowledge.", "routing_description": "Use only when the user explicitly asks to remember or save a chat message.", "routing_examples": ["Remember this answer"], "estimated_cost_class": "low", "required_capabilities": ["knowledge_write"], "graph": REMEMBER},
     {"slug": "save-source", "name": "Save Source", "description": "Save a website snapshot to knowledge.", "routing_description": "Use when the user explicitly asks to save a URL as a knowledge source.", "routing_examples": ["Save this website to my database"], "estimated_cost_class": "low", "required_capabilities": ["web", "knowledge_write"], "graph": SAVE_SOURCE},
     {"slug": "research", "name": "Research", "description": "Bounded two-round evidence research with source validation.", "routing_description": "Choose only when the standalone request needs an investigation rather than a direct lookup: compare sources, resolve uncertainty, validate claims, synthesize several angles, or perform a second search if first-pass evidence is insufficient. Do not choose this solely because web evidence is needed, because the topic is important, or because information is current. Choose Web Answer for direct web-grounded questions.", "routing_examples": ["Research the competing explanations and cite sources", "Compare the strongest evidence for both claims", "Investigate this topic in depth"], "estimated_cost_class": "high", "required_capabilities": ["web", "chat"], "graph": RESEARCH},