docs/docs/configuration/reference.md

@@ -911,7 +911,7 @@ cameras:
       user: admin
       # Optional: password for login.
       password: admin
-      # Optional: Skip TLS verification and disable digest authentication for the ONVIF server (default: shown below)
+      # Optional: Skip TLS verification from the ONVIF server (default: shown below)
       tls_insecure: False
       # Optional: Ignores time synchronization mismatches between the camera and the server during authentication.
       # Using NTP on both ends is recommended and this should only be set to True in a "safe" environment due to the security risk it represents.

frigate/data_processing/post/review_descriptions.py

@@ -251,22 +251,20 @@ class ReviewDescriptionProcessor(PostProcessorApi):
             if not primary_segments:
                 return "No concerns were found during this time period."
 
-            # Build hierarchical structure: each primary event with its contextual items
+            # For each primary segment, find overlapping contextual items from other cameras
-            events_with_context = []
+            all_items_for_summary = []
 
             for primary_seg in primary_segments:
-                # Start building the primary event structure
+                # Add the primary item with marker
                 primary_item = copy.deepcopy(primary_seg["metadata"])
-                primary_item["camera"] = primary_seg["camera"]
+                primary_item["_is_primary"] = True
-                primary_item["start_time"] = primary_seg["start_time"]
+                primary_item["_camera"] = primary_seg["camera"]
-                primary_item["end_time"] = primary_seg["end_time"]
+                all_items_for_summary.append(primary_item)
 
                 # Find overlapping contextual items from other cameras
                 primary_start = primary_seg["start_time"]
                 primary_end = primary_seg["end_time"]
                 primary_camera = primary_seg["camera"]
-                contextual_items = []
-                seen_contextual_cameras = set()
 
                 for seg in segments:
                     seg_camera = seg["camera"]
@@ -281,25 +279,21 @@ class ReviewDescriptionProcessor(PostProcessorApi):
                     seg_end = seg["end_time"]
 
                     if seg_start < primary_end and primary_start < seg_end:
-                        # Avoid duplicates if same camera has multiple overlapping segments
-                        if seg_camera not in seen_contextual_cameras:
                         contextual_item = copy.deepcopy(seg["metadata"])
-                            contextual_item["camera"] = seg_camera
+                        contextual_item["_is_primary"] = False
-                            contextual_item["start_time"] = seg_start
+                        contextual_item["_camera"] = seg_camera
-                            contextual_item["end_time"] = seg_end
+                        contextual_item["_related_to_camera"] = primary_camera
-                            contextual_items.append(contextual_item)
-                            seen_contextual_cameras.add(seg_camera)
 
-                # Add context array to primary item
+                        if not any(
-                primary_item["context"] = contextual_items
+                            item.get("_camera") == seg_camera
-                events_with_context.append(primary_item)
+                            and item.get("time") == contextual_item.get("time")
+                            for item in all_items_for_summary
+                        ):
+                            all_items_for_summary.append(contextual_item)
 
-            total_context_items = sum(
-                len(event.get("context", [])) for event in events_with_context
-            )
             logger.debug(
-                f"Summary includes {len(events_with_context)} primary events with "
+                f"Summary includes {len(primary_segments)} primary items and "
-                f"{total_context_items} total contextual items"
+                f"{len(all_items_for_summary) - len(primary_segments)} contextual items"
             )
 
             if self.config.review.genai.debug_save_thumbnails:
@@ -310,7 +304,7 @@ class ReviewDescriptionProcessor(PostProcessorApi):
             return self.genai_client.generate_review_summary(
                 start_ts,
                 end_ts,
-                events_with_context,
+                all_items_for_summary,
                 self.config.review.genai.debug_save_thumbnails,
             )
         else:

frigate/genai/__init__.py

@@ -177,60 +177,78 @@ Each line represents a detection state, not necessarily unique individuals. Pare
         self,
         start_ts: float,
         end_ts: float,
-        events: list[dict[str, Any]],
+        segments: list[dict[str, Any]],
         debug_save: bool,
     ) -> str | None:
         """Generate a summary of review item descriptions over a period of time."""
         time_range = f"{datetime.datetime.fromtimestamp(start_ts).strftime('%B %d, %Y at %I:%M %p')} to {datetime.datetime.fromtimestamp(end_ts).strftime('%B %d, %Y at %I:%M %p')}"
         timeline_summary_prompt = f"""
-You are a security officer writing a concise security report.
+You are a security officer.
+Time range: {time_range}.
+Input: JSON list with "title", "scene", "confidence", "potential_threat_level" (0-2), "other_concerns", "_is_primary", "_camera".
 
-Time range: {time_range}
+Task: Write a concise, human-presentable security report in markdown format.
 
-Input format: Each event is a JSON object with:
+CRITICAL - Understanding Primary vs Contextual Items:
-- "title", "scene", "confidence", "potential_threat_level" (0-2), "other_concerns", "camera", "time", "start_time", "end_time"
+- Items with "_is_primary": true are events that REQUIRE REVIEW and MUST be included in the report
-- "context": array of related events from other cameras that occurred during overlapping time periods
+- Items with "_is_primary": false are additional context from other camera perspectives that overlap in time
+- **DO NOT create separate bullet points or sections for contextual items**
+- **ONLY use contextual items to enrich and inform the description of primary items**
+- The "_camera" field indicates which camera captured each event
+- **When a contextual item provides relevant background, you MUST incorporate it directly into the primary event's bullet point**
+- Contextual information often explains or de-escalates seemingly suspicious primary events
 
-Report Structure - Use this EXACT format:
+Rules for the report:
 
+- Title & overview
+  - Start with:
     # Security Summary - {time_range}
+  - Write a 1-2 sentence situational overview capturing the general pattern of the period.
+  - Keep the overview high-level; specific details will be in the event bullets below.
 
-## Overview
+- Event details
-[Write 1-2 sentences summarizing the overall activity pattern during this period.]
+  - **ONLY create bullet points for PRIMARY items (_is_primary: true)**
+  - **Do NOT create sections or bullets for events that don't exist**
+  - Do NOT create separate bullets for contextual items
+  - Present primary events in chronological order as a bullet list.
+  - **CRITICAL: When contextual items overlap with a primary event, you MUST weave that information directly into the same bullet point**
+  - Format: **[Timestamp]** - [Description incorporating any contextual information]. [Camera info]. (threat level: X)
+  - If contextual information provides an explanation (e.g., delivery truck → person is likely delivery driver), reflect this understanding in your description and potentially adjust the perceived threat level
+  - If multiple PRIMARY events occur within the same minute, combine them into a single bullet with sub-points.
+  - Use bold timestamps for clarity.
+  - Camera format: "Camera: [camera name]" or mention contextual cameras inline when relevant
+  - Group bullets under subheadings ONLY when you have actual PRIMARY events to list (e.g., Porch Activity, Unusual Behavior).
 
----
+- Threat levels
+  - Show the threat level for PRIMARY events using these labels:
+    - Threat level 0: "Normal"
+    - Threat level 1: "Needs review"
+    - Threat level 2: "Security concern"
+  - Format as (threat level: Normal), (threat level: Needs review), or (threat level: Security concern).
+  - **When contextual items clearly explain a primary event (e.g., delivery truck explains person at door), you should describe it as normal activity and note the explanation**
+  - **Your description and tone should reflect the fuller understanding provided by contextual information**
+  - Example: Primary event says "unidentified person with face covering" but context shows delivery truck → describe as "delivery person (truck visible on Front Driveway Cam)" rather than emphasizing suspicious elements
+  - The stored threat level remains as originally classified, but your narrative should reflect the contextual understanding
+  - If multiple PRIMARY events at the same time share the same threat level, only state it once.
 
-## Timeline
+- Final assessment
+  - End with a Final Assessment section.
+  - If all primary events are threat level 0 or explained by contextual items:
+    Final assessment: Only normal residential activity observed during this period.
+  - If threat level 1 events are present:
+    Final assessment: Some activity requires review but no security concerns identified.
+  - If threat level 2 events are present, clearly summarize them as Security concerns requiring immediate attention.
+  - Keep this section brief - do not repeat details from the event descriptions above.
 
-[Group events by time periods (e.g., "Morning (6:00 AM - 12:00 PM)", "Afternoon (12:00 PM - 5:00 PM)", "Evening (5:00 PM - 9:00 PM)", "Night (9:00 PM - 6:00 AM)"). Use appropriate time blocks based on when events occurred.]
+- Conciseness
-
+  - Do not repeat benign clothing/appearance details unless they distinguish individuals.
-### [Time Block Name]
+  - Summarize similar routine events instead of restating full scene descriptions.
-
+  - When incorporating contextual information, do so briefly and naturally within the primary event description.
-**HH:MM AM/PM** | [Camera Name] | [Threat Level Indicator]
+  - Avoid lengthy explanatory notes - integrate context seamlessly into the narrative.
-- [Event title]: [Clear description incorporating contextual information from the "context" array]
-- Context: [If context array has items, mention them here, e.g., "Delivery truck present on Front Driveway Cam (HH:MM AM/PM)"]
-- Assessment: [Brief assessment incorporating context - if context explains the event, note it here]
-
-[Repeat for each event in chronological order within the time block]
-
----
-
-## Summary
-[One sentence summarizing the period. If all events are normal/explained: "Routine activity observed." If review needed: "Some activity requires review but no security concerns." If security concerns: "Security concerns requiring immediate attention."]
-
-Guidelines:
-- List ALL events in chronological order, grouped by time blocks
-- Threat level indicators: ✓ Normal, ⚠️ Needs review, 🔴 Security concern
-- Integrate contextual information naturally - use the "context" array to enrich each event's description
-- If context explains the event (e.g., delivery truck explains person at door), describe it accordingly (e.g., "delivery person" not "unidentified person")
-- Be concise but informative - focus on what happened and what it means
-- If contextual information makes an event clearly normal, reflect that in your assessment
-- Only create time blocks that have events - don't create empty sections
 """
 
-        timeline_summary_prompt += "\n\nEvents:\n"
+        for item in segments:
-        for event in events:
+            timeline_summary_prompt += f"\n{item}"
-            timeline_summary_prompt += f"\n{event}\n"
 
         if debug_save:
             with open(