fix-generate-summary-5.1 - Fixed 5.1 conversation summary error by having the summary code use the same multi-endpoint model resolution that chat uses.

application/single_app/route_backend_conversation_export.py

@@ -9,7 +9,7 @@
 from collections import Counter, defaultdict
 from datetime import datetime
 from html import escape as _escape_html
-from typing import Any, Dict, List, Optional
+from typing import Any, Dict, List, Optional, Tuple
 
 from bs4 import BeautifulSoup, NavigableString, Tag
 from config import *
@@ -56,6 +56,9 @@ def api_export_conversations():
             packaging (str): Output packaging — "single" or "zip".
             include_summary_intro (bool): Whether to generate a per-conversation intro.
             summary_model_deployment (str): Optional model deployment for summary generation.
+            summary_model_id (str): Optional multi-endpoint model id for summary generation.
+            summary_model_endpoint_id (str): Optional multi-endpoint endpoint id for summary generation.
+            summary_model_provider (str): Optional provider (aoai/aifoundry) for summary generation.
         """
         user_id = get_current_user_id()
         if not user_id:
@@ -70,6 +73,9 @@ def api_export_conversations():
         packaging = str(data.get('packaging', 'single')).lower()
         include_summary_intro = bool(data.get('include_summary_intro', False))
         summary_model_deployment = str(data.get('summary_model_deployment', '') or '').strip()
+        summary_model_id = str(data.get('summary_model_id', '') or '').strip()
+        summary_model_endpoint_id = str(data.get('summary_model_endpoint_id', '') or '').strip()
+        summary_model_provider = str(data.get('summary_model_provider', '') or '').strip()
 
         if not conversation_ids or not isinstance(conversation_ids, list):
             return jsonify({'error': 'At least one conversation_id is required'}), 400
@@ -115,7 +121,10 @@ def api_export_conversations():
                         user_id=user_id,
                         settings=settings,
                         include_summary_intro=include_summary_intro,
-                        summary_model_deployment=summary_model_deployment
+                        summary_model_deployment=summary_model_deployment,
+                        summary_model_id=summary_model_id,
+                        summary_model_endpoint_id=summary_model_endpoint_id,
+                        summary_model_provider=summary_model_provider
                     )
                 )
 
@@ -247,7 +256,10 @@ def _build_export_entry(
     user_id: str,
     settings: Dict[str, Any],
     include_summary_intro: bool = False,
-    summary_model_deployment: str = ''
+    summary_model_deployment: str = '',
+    summary_model_id: str = '',
+    summary_model_endpoint_id: str = '',
+    summary_model_provider: str = '',
 ) -> Dict[str, Any]:
     artifact_payload_map = build_message_artifact_payload_map(raw_messages)
     filtered_messages = _filter_messages_for_export(raw_messages)
@@ -309,6 +321,9 @@ def _build_export_entry(
         settings=settings,
         enabled=include_summary_intro,
         summary_model_deployment=summary_model_deployment,
+        summary_model_id=summary_model_id,
+        summary_model_endpoint_id=summary_model_endpoint_id,
+        summary_model_provider=summary_model_provider,
         message_time_start=message_time_start,
         message_time_end=message_time_end
     )
@@ -660,7 +675,12 @@ def generate_conversation_summary(
     model_deployment: str,
     message_time_start: str = None,
     message_time_end: str = None,
-    conversation_id: str = None
+    conversation_id: str = None,
+    model_id: Optional[str] = None,
+    model_endpoint_id: Optional[str] = None,
+    model_provider: Optional[str] = None,
+    user_id: Optional[str] = None,
+    active_group_ids: Optional[List[str]] = None,
 ) -> Dict[str, Any]:
     """Generate a conversation summary using the LLM and optionally persist it.
 
@@ -686,7 +706,25 @@ def generate_conversation_summary(
 
     transcript_text = _truncate_for_summary(transcript_text)
 
-    gpt_client, gpt_model = _initialize_gpt_client(settings, model_deployment)
+    # Prefer multi-endpoint model resolution so summaries use the same
+    # AOAI endpoint/deployment as chat for a given model selection.
+    gpt_client: AzureOpenAI
+    gpt_model: str
+
+    multi_endpoint_client = _initialize_summary_multi_endpoint_client(
+        settings,
+        model_deployment=model_deployment,
+        model_id=model_id,
+        model_endpoint_id=model_endpoint_id,
+        model_provider=model_provider,
+        user_id=user_id,
+        active_group_ids=active_group_ids or [],
+    )
+
+    if multi_endpoint_client is not None:
+        gpt_client, gpt_model = multi_endpoint_client
+    else:
+        gpt_client, gpt_model = _initialize_gpt_client(settings, model_deployment)
     summary_prompt = (
         "You are summarizing a conversation for an export document. "
         "Read the full conversation below and write a concise summary. "
@@ -761,6 +799,9 @@ def _build_summary_intro(
     settings: Dict[str, Any],
     enabled: bool,
     summary_model_deployment: str,
+    summary_model_id: str = '',
+    summary_model_endpoint_id: str = '',
+    summary_model_provider: str = '',
     message_time_start: str = None,
     message_time_end: str = None
 ) -> Dict[str, Any]:
@@ -809,7 +850,12 @@ def _build_summary_intro(
             model_deployment=summary_model_deployment,
             message_time_start=message_time_start,
             message_time_end=message_time_end,
-            conversation_id=conversation_id
+            conversation_id=conversation_id,
+            model_id=summary_model_id or None,
+            model_endpoint_id=summary_model_endpoint_id or None,
+            model_provider=summary_model_provider or None,
+            user_id=conversation.get('user_id'),
+            active_group_ids=[],
         )
 
         summary_intro.update({
@@ -848,6 +894,55 @@ def _truncate_for_summary(transcript_text: str) -> str:
     )
 
 
+def _initialize_summary_multi_endpoint_client(
+    settings: Dict[str, Any],
+    model_deployment: str,
+    model_id: Optional[str] = None,
+    model_endpoint_id: Optional[str] = None,
+    model_provider: Optional[str] = None,
+    user_id: Optional[str] = None,
+    active_group_ids: Optional[List[str]] = None,
+) -> Optional[Tuple[AzureOpenAI, str]]:
+    """Initialize a GPT client for summaries using the multi-endpoint model config.
+
+    This mirrors the streaming model resolution path so that summaries use the
+    exact same AOAI endpoint/deployment as chat for a given model selection.
+    Returns (client, deployment) when a multi-endpoint config can be resolved,
+    otherwise None to allow fallback to the legacy single-endpoint settings.
+    """
+    try:
+        from route_backend_chats import resolve_streaming_multi_endpoint_gpt_config
+    except Exception:
+        return None
+
+    if not settings.get('enable_multi_model_endpoints', False):
+        return None
+
+    request_payload = {
+        'model_deployment': model_deployment or '',
+        'model_id': model_id or '',
+        'model_endpoint_id': model_endpoint_id or '',
+        'model_provider': model_provider or '',
+    }
+
+    try:
+        resolved = resolve_streaming_multi_endpoint_gpt_config(
+            settings,
+            request_payload,
+            user_id or '',
+            active_group_ids=active_group_ids or [],
+            allow_default_selection=False,
+        )
+    except Exception:
+        return None
+
+    if not resolved:
+        return None
+
+    gpt_client, deployment, _provider, _endpoint, _auth_settings, _api_version = resolved
+    return gpt_client, deployment
+
+
 def _initialize_gpt_client(settings: Dict[str, Any], requested_model: str = ''):
     enable_gpt_apim = settings.get('enable_gpt_apim', False)
 

application/single_app/route_backend_conversations.py

@@ -1000,6 +1000,9 @@ def generate_conversation_summary_api(conversation_id):
 
         body = request.get_json(silent=True) or {}
         model_deployment = body.get('model_deployment', '')
+        model_id = body.get('model_id') or None
+        model_endpoint_id = body.get('model_endpoint_id') or None
+        model_provider = body.get('model_provider') or None
 
         # Query messages for this conversation
         try:
@@ -1045,7 +1048,17 @@ def generate_conversation_summary_api(conversation_id):
                 model_deployment=model_deployment,
                 message_time_start=message_time_start,
                 message_time_end=message_time_end,
-                conversation_id=conversation_id
+                conversation_id=conversation_id,
+                model_id=model_id,
+                model_endpoint_id=model_endpoint_id,
+                model_provider=model_provider,
+                user_id=user_id,
+                # For now, summaries use the same group context that
+                # was already captured in conversation metadata. The
+                # generate_conversation_summary helper accepts
+                # active_group_ids but does not require it, so we pass
+                # None here to avoid referencing undefined variables.
+                active_group_ids=None,
             )
             return jsonify({'success': True, 'summary': summary_data}), 200
 

application/single_app/static/js/chat/chat-conversation-details.js

@@ -673,19 +673,43 @@ function getAvailableModelOptions() {
   if (!globalSelect) {
     return '<option value="">Default</option>';
   }
-  let options = '';
+
+  const tempSelect = document.createElement('select');
+
   for (const opt of globalSelect.options) {
-    options += `<option value="${escapeHtml(opt.value)}"${opt.selected ? ' selected' : ''}>${escapeHtml(opt.text)}</option>`;
+    const cloned = document.createElement('option');
+    cloned.value = opt.value;
+    cloned.textContent = opt.text;
+
+    if (opt.dataset?.deploymentName) {
+      cloned.dataset.deploymentName = opt.dataset.deploymentName;
+    }
+    if (opt.dataset?.modelId) {
+      cloned.dataset.modelId = opt.dataset.modelId;
+    }
+    if (opt.dataset?.endpointId) {
+      cloned.dataset.endpointId = opt.dataset.endpointId;
+    }
+    if (opt.dataset?.provider) {
+      cloned.dataset.provider = opt.dataset.provider;
+    }
+    if (opt.selected) {
+      cloned.selected = true;
+    }
+
+    tempSelect.appendChild(cloned);
   }
-  return options || '<option value="">Default</option>';
+
+  const html = tempSelect.innerHTML;
+  return html || '<option value="">Default</option>';
 }
 
 /**
  * Handle summary generation (generate or regenerate)
  * @param {string} conversationId - The conversation ID
- * @param {string} modelDeployment - Selected model deployment
+ * @param {Object} modelPayload - Selected model metadata
  */
-async function handleGenerateSummary(conversationId, modelDeployment) {
+async function handleGenerateSummary(conversationId, modelPayload) {
   const cardBody = document.getElementById('summary-card-body');
   if (!cardBody) {
     return;
@@ -704,7 +728,7 @@ async function handleGenerateSummary(conversationId, modelDeployment) {
     const response = await fetch(`/api/conversations/${conversationId}/summary`, {
       method: 'POST',
       headers: { 'Content-Type': 'application/json' },
-      body: JSON.stringify({ model_deployment: modelDeployment })
+      body: JSON.stringify(modelPayload || {})
     });
 
     if (!response.ok) {
@@ -766,8 +790,18 @@ document.addEventListener('click', function(e) {
     const cid = btn.getAttribute('data-conversation-id');
     const modelSelect = document.getElementById('summary-model-select');
     const selectedOption = modelSelect ? modelSelect.options[modelSelect.selectedIndex] : null;
-    const model = selectedOption?.dataset?.deploymentName || (modelSelect ? modelSelect.value : '');
-    handleGenerateSummary(cid, model);
+
+    const modelDeployment = selectedOption?.dataset?.deploymentName || (modelSelect ? modelSelect.value : '');
+    const modelId = selectedOption?.dataset?.modelId || '';
+    const modelEndpointId = selectedOption?.dataset?.endpointId || '';
+    const modelProvider = selectedOption?.dataset?.provider || '';
+
+    handleGenerateSummary(cid, {
+      model_deployment: modelDeployment,
+      model_id: modelId,
+      model_endpoint_id: modelEndpointId,
+      model_provider: modelProvider
+    });
     return;
   }
 
@@ -779,8 +813,17 @@ document.addEventListener('click', function(e) {
     // Use the currently selected global model for regeneration
     const globalSelect = document.getElementById('model-select');
     const selectedOption = globalSelect ? globalSelect.options[globalSelect.selectedIndex] : null;
-    const model = selectedOption?.dataset?.deploymentName || (globalSelect ? globalSelect.value : '');
-    handleGenerateSummary(cid, model);
+    const modelDeployment = selectedOption?.dataset?.deploymentName || (globalSelect ? globalSelect.value : '');
+    const modelId = selectedOption?.dataset?.modelId || '';
+    const modelEndpointId = selectedOption?.dataset?.endpointId || '';
+    const modelProvider = selectedOption?.dataset?.provider || '';
+
+    handleGenerateSummary(cid, {
+      model_deployment: modelDeployment,
+      model_id: modelId,
+      model_endpoint_id: modelEndpointId,
+      model_provider: modelProvider
+    });
     return;
   }
 

application/single_app/static/js/chat/chat-export.js

@@ -17,6 +17,9 @@ let exportFormat = 'json';
 let exportPackaging = 'single';
 let includeSummaryIntro = false;
 let summaryModelDeployment = '';
+let summaryModelId = '';
+let summaryModelEndpointId = '';
+let summaryModelProvider = '';
 let currentStep = 1;
 let totalSteps = 3;
 let skipSelectionStep = false;
@@ -57,6 +60,10 @@ function openExportWizard(conversationIds, skipSelection) {
     exportPackaging = conversationIds.length > 1 ? 'zip' : 'single';
     includeSummaryIntro = false;
     summaryModelDeployment = _getDefaultSummaryModel();
+    const defaultModelMeta = _getDefaultSummaryModelMetadata();
+    summaryModelId = defaultModelMeta.modelId;
+    summaryModelEndpointId = defaultModelMeta.endpointId;
+    summaryModelProvider = defaultModelMeta.provider;
     skipSelectionStep = !!skipSelection;
 
     // Determine step configuration
@@ -345,9 +352,17 @@ function _renderSummaryStep(container) {
 
     if (summaryModelSelect && hasModelOptions) {
         summaryModelSelect.value = defaultSummaryModel || summaryModelSelect.value;
-        summaryModelDeployment = summaryModelSelect.options[summaryModelSelect.selectedIndex]?.dataset?.deploymentName || summaryModelSelect.value;
+        const selected = summaryModelSelect.options[summaryModelSelect.selectedIndex];
+        summaryModelDeployment = selected?.dataset?.deploymentName || summaryModelSelect.value;
+        summaryModelId = selected?.dataset?.modelId || '';
+        summaryModelEndpointId = selected?.dataset?.endpointId || '';
+        summaryModelProvider = selected?.dataset?.provider || '';
         summaryModelSelect.addEventListener('change', () => {
-            summaryModelDeployment = summaryModelSelect.options[summaryModelSelect.selectedIndex]?.dataset?.deploymentName || summaryModelSelect.value;
+            const opt = summaryModelSelect.options[summaryModelSelect.selectedIndex];
+            summaryModelDeployment = opt?.dataset?.deploymentName || summaryModelSelect.value;
+            summaryModelId = opt?.dataset?.modelId || '';
+            summaryModelEndpointId = opt?.dataset?.endpointId || '';
+            summaryModelProvider = opt?.dataset?.provider || '';
         });
     }
 
@@ -359,7 +374,11 @@ function _renderSummaryStep(container) {
             }
             if (includeSummaryIntro && summaryModelSelect && !summaryModelSelect.value) {
                 summaryModelSelect.value = _getDefaultSummaryModel();
-                summaryModelDeployment = summaryModelSelect.options[summaryModelSelect.selectedIndex]?.dataset?.deploymentName || summaryModelSelect.value;
+                const opt = summaryModelSelect.options[summaryModelSelect.selectedIndex];
+                summaryModelDeployment = opt?.dataset?.deploymentName || summaryModelSelect.value;
+                summaryModelId = opt?.dataset?.modelId || '';
+                summaryModelEndpointId = opt?.dataset?.endpointId || '';
+                summaryModelProvider = opt?.dataset?.provider || '';
             }
         });
     }
@@ -532,7 +551,10 @@ async function _executeExport() {
                 format: exportFormat,
                 packaging: exportPackaging,
                 include_summary_intro: includeSummaryIntro,
-                summary_model_deployment: includeSummaryIntro ? summaryModelDeployment : null
+                summary_model_deployment: includeSummaryIntro ? summaryModelDeployment : null,
+                summary_model_id: includeSummaryIntro ? summaryModelId : null,
+                summary_model_endpoint_id: includeSummaryIntro ? summaryModelEndpointId : null,
+                summary_model_provider: includeSummaryIntro ? summaryModelProvider : null
             })
         });
 
@@ -605,6 +627,20 @@ function _getDefaultSummaryModel() {
     return mainModelSelect.value || (mainModelSelect.options[0] ? mainModelSelect.options[0].value : '');
 }
 
+function _getDefaultSummaryModelMetadata() {
+    const mainModelSelect = getEl('model-select');
+    if (!mainModelSelect || !mainModelSelect.options.length) {
+        return { modelId: '', endpointId: '', provider: '' };
+    }
+
+    const selected = mainModelSelect.options[mainModelSelect.selectedIndex] || mainModelSelect.options[0];
+    return {
+        modelId: selected.dataset?.modelId || '',
+        endpointId: selected.dataset?.endpointId || '',
+        provider: selected.dataset?.provider || ''
+    };
+}
+
 // --- Expose Globally ---
 window.chatExport = {
     openExportWizard