Jobs: forward cached GPT results, reduce stream buffer to 60, wire up first_frame_gpt_results

app.py

@@ -405,7 +405,7 @@ async def detect_async_endpoint(
     active_depth = depth_estimator if enable_depth else None
 
     try:
-        processed_frame, detections, depth_map = process_first_frame(
+        processed_frame, detections, depth_map, first_frame_gpt_results = process_first_frame(
             str(input_path),
             query_list,
             mode=mode,
@@ -455,6 +455,7 @@ async def detect_async_endpoint(
         enable_gpt=enable_gpt,
         mission_spec=mission_spec,
         mission_mode=mission_mode,
+        first_frame_gpt_results=first_frame_gpt_results,
     )
     get_job_storage().create(job)
     asyncio.create_task(process_video_async(job_id))

jobs/background.py

@@ -53,6 +53,7 @@ async def process_video_async(job_id: str) -> None:
                 job.enable_gpt,
                 stream_queue,
                 job.mission_spec,  # Forward mission spec for relevance gating
+                job.first_frame_gpt_results,  # Avoid duplicate GPT call on frame 0
             )
             detection_path, detections_list = result_pkg
 

jobs/models.py

@@ -37,3 +37,4 @@ class JobInfo:
     # Mission specification (None = LEGACY mode)
     mission_spec: Optional[Any] = None  # utils.schemas.MissionSpecification
     mission_mode: str = "LEGACY"  # "MISSION" or "LEGACY"
+    first_frame_gpt_results: Optional[Dict[str, Any]] = None  # Cached GPT results from process_first_frame

jobs/streaming.py

@@ -13,7 +13,7 @@ def create_stream(job_id: str) -> queue.Queue:
         # standard Queue, thread-safe
         # maxsize to prevent memory explosion if consumer is slow
         # Buffer increased to 300 (approx 10s at 30fps) for smooth streaming
-        q = queue.Queue(maxsize=300) 
+        q = queue.Queue(maxsize=60)
         _STREAMS[job_id] = q
         return q