deploy: batch update 1 file(s)

deploy: batch update 1 file(s)

deploy: batch update 2 file(s)

deploy: batch update 2 file(s)

deploy: batch update 3 file(s)

deploy: batch update 2 file(s)

deploy: batch update 1 file(s)

deploy: batch update 1 file(s)

deploy: batch update 1 file(s)

deploy: batch update 1 file(s)

deploy: batch update 1 file(s)

deploy: batch update 2 file(s)

deploy: batch update 2 file(s)

deploy: batch update 1 file(s)

deploy: batch update 2 file(s)

deploy: batch update 5 file(s)

chore: trigger rebuild from paused state

revert: Dockerfile back to python app.py (gunicorn caused stuck)

deploy: batch update 3 file(s)

deploy: batch update 2 file(s)

deploy: batch update 1 file(s)

deploy: batch update 1 file(s)

deploy: batch update 5 file(s)

fix: deferred heavy imports — Flask starts in <1min on cpu-basic

debug: full app.py with startup debug prints

debug: minimal app.py to test startup

update Dockerfile

fix: defer heavy imports to background thread — Flask starts in <2s

fix: background model loading so Flask starts instantly (fixes restart loop)

deploy: batch update - alerts fix + readiness reset

Upload behavior_analysis/action_recognition.py with huggingface_hub

Upload real_time_alerts.py with huggingface_hub

Upload live_stream_processor.py with huggingface_hub

Upload live_stream_processor.py with huggingface_hub

Upload report_generation/report_builder.py with huggingface_hub

Upload app.py with huggingface_hub

Upload live_stream_processor.py with huggingface_hub

Fix webcam browser mode, face dedup by person, confidence field, video proxy, upsert faces

fix: browser webcam, face dedup in reports, person clip timestamps

fix: keyframe images, video clips, evidence images, live stream webcam+URL, remove demo mode

fix: filter embeddings by dimension to prevent np.stack shape mismatch in search

fix: search always 503, embeddings not in MongoDB, report gen imports

fix: remove duplicate health_check route (AssertionError on startup)

fix: remove llama-cpp-python (C++ build times out on free tier)

fix: prevent CUDA torch re-download and speed up llama-cpp-python build

fix: replace removed Debian trixie packages (libgl1-mesa-glx -> libgl1, libgdk-pixbuf2.0-0 -> libgdk-pixbuf-xlib-2.0-0)

fix: use JSON array form for COPY with spaces in filename

feat: Full DetectifAI backend with B2 storage, DEMO_MODE, Stripe bypass

initial commit