Fix: Use test query ONCE for readiness, detects when docs actually searchable

src/notebookllama/pages/3_Notebook_Chat.py

@@ -101,7 +101,7 @@ if not documents:
 # Check if pipeline is actually ready
 from background_tasks import get_notebook_processing_tasks, TaskStatus
 from document_manager import get_latest_document_summary
-from pipeline_manager import check_pipeline_status_direct
+from pipeline_manager import check_pipeline_ready_with_test
 import asyncio as aio
 import time
 
@@ -138,8 +138,8 @@ if f'pipeline_confirmed_ready_{notebook.id}' not in st.session_state:
 
             st.stop()
 
-        with st.spinner("🔍 Checking pipeline status..."):
-            readiness = asyncio.run(check_pipeline_status_direct(notebook.pipeline_id))
+        with st.spinner("🔍 Checking if documents are searchable..."):
+            readiness = asyncio.run(check_pipeline_ready_with_test(notebook.pipeline_id, notebook.model_type))
 
             # If ready, mark as confirmed and don't test again
             if readiness.get('ready', False):

src/notebookllama/pipeline_manager.py

@@ -85,53 +85,55 @@ async def add_document_to_pipeline(pipeline_id: str, file_path: str) -> bool:
         return False
 
 
-async def check_pipeline_status_direct(pipeline_id: str) -> dict:
+async def check_pipeline_ready_with_test(pipeline_id: str, model_type: str) -> dict:
     """
-    Check pipeline status directly from LlamaCloud API (no test queries)
+    Check if pipeline is actually ready by doing ONE simple test query
+    Only called during readiness check, not during chat
 
     Returns:
-        dict with 'ready' (bool), 'status' (str), 'doc_count' (int), 'indexed_count' (int)
+        dict with 'ready' (bool), 'status' (str), 'message' (str)
     """
     try:
+        # First check if pipeline has documents
         client = AsyncLlamaCloud(token=os.getenv("LLAMACLOUD_API_KEY"))
-
-        # Get pipeline details
         pipeline = await client.pipelines.get_pipeline(pipeline_id=pipeline_id)
 
-        # Count documents in pipeline
         doc_count = 0
-        indexed_count = 0
+        if hasattr(pipeline, 'configured_transformation') and hasattr(pipeline.configured_transformation, 'data_sources'):
+            doc_count = len(pipeline.configured_transformation.data_sources)
 
-        if hasattr(pipeline, 'configured_transformation'):
-            transform = pipeline.configured_transformation
-            if hasattr(transform, 'data_sources'):
-                doc_count = len(transform.data_sources)
+        if doc_count == 0:
+            return {
+                'ready': False,
+                'status': 'No documents',
+                'message': 'No documents in pipeline yet'
+            }
 
-                # Check each document's status
-                for source in transform.data_sources:
-                    # If source has been processed, it's indexed
-                    # LlamaCloud doesn't expose explicit "indexed" status,
-                    # so we assume if it's in data_sources, it's processed
-                    indexed_count += 1
+        # Documents exist - now test if they're actually searchable
+        print(f"Testing pipeline {pipeline_id} with simple query...")
+        test_response = await query_notebook_pipeline(pipeline_id, "Summarize this document briefly", model_type)
 
-        # Pipeline is ready if it has documents
-        is_ready = doc_count > 0 and indexed_count > 0
-
-        return {
-            'ready': is_ready,
-            'status': 'Ready' if is_ready else 'Indexing',
-            'doc_count': doc_count,
-            'indexed_count': indexed_count,
-            'message': f'{indexed_count}/{doc_count} documents indexed' if doc_count > 0 else 'No documents in pipeline'
-        }
+        # Check if we got a real response
+        if test_response and "Empty Response" not in test_response and "Error:" not in test_response and len(test_response) > 50:
+            print(f"✓ Pipeline ready - got real response")
+            return {
+                'ready': True,
+                'status': 'Ready',
+                'message': f'{doc_count} document(s) indexed and searchable'
+            }
+        else:
+            print(f"⏳ Pipeline not ready yet - got: {test_response[:100] if test_response else 'None'}")
+            return {
+                'ready': False,
+                'status': 'Still indexing',
+                'message': f'{doc_count} document(s) added but not searchable yet'
+            }
 
     except Exception as e:
-        print(f"Error checking pipeline status: {e}")
+        print(f"Error checking readiness: {e}")
         return {
             'ready': False,
             'status': 'Error',
-            'doc_count': 0,
-            'indexed_count': 0,
             'message': f'Error: {str(e)[:100]}'
         }