Comprehensive configuration reference for Streamware after the recent refactoring.
Streamware has been fully refactored to eliminate hardcoded values and provide complete configurability through environment variables and .env files. All detection thresholds, timeouts, and model parameters can now be customized without code modifications.
.envAll configuration is managed through environment variables. The system automatically loads from .env file in the project root.
config.pyDefault values are defined in streamware/config.py for reference and fallback.
# Detection sensitivity (lower = more sensitive)
SQ_YOLO_CONFIDENCE_THRESHOLD=0.15 # YOLO detection confidence threshold
SQ_YOLO_CONFIDENCE_THRESHOLD_HIGH=0.3 # Higher threshold for reducing false positives
Impact:
# HOG detection parameters
SQ_HOG_SCALE=400 # Max dimension for HOG detection
SQ_HOG_WINSTRIDE=8 # HOG detection window stride
SQ_HOG_PADDING=4 # HOG detection padding
SQ_HOG_SCALE_FACTOR=1.05 # HOG detection scale factor
SQ_HOG_CONFIDENT_NO_PERSON=0.8 # Confidence when no person detected
SQ_HOG_MIGHT_BE_PERSON=0.5 # Confidence when might be person
# Motion detection thresholds
SQ_MOTION_DIFF_THRESHOLD=25 # OpenCV threshold for motion detection
SQ_MOTION_BLUR_KERNEL=5 # Gaussian blur kernel size
SQ_MOTION_CONTOUR_MIN_AREA=100 # Minimum contour area for motion regions
SQ_MOTION_SCALE_WIDTH=320 # Max width for motion detection scaling
SQ_MOTION_SCALE_HEIGHT=240 # Max height for motion detection scaling
SQ_MOTION_FIRST_FRAME_PERCENT=100.0 # Assume motion percent for first frame
SQ_MOTION_ERROR_PERCENT=50.0 # Motion error handling
# Motion classification thresholds
SQ_MOTION_MIN_PERCENT=0.5 # Minimum motion percent to consider
SQ_MOTION_LOW_PERCENT=1.0 # Low motion threshold
SQ_MOTION_MEDIUM_PERCENT=5.0 # Medium motion threshold
SQ_MOTION_HIGH_PERCENT=15.0 # High motion threshold
# Timeout values for various LLM operations (in seconds)
SQ_GUARDER_TIMEOUT=5 # Timeout for guarder model availability check
SQ_QUICK_PERSON_TIMEOUT=10 # Timeout for quick person detection
SQ_QUICK_CHANGE_TIMEOUT=8 # Timeout for quick change detection
SQ_SUMMARIZE_TIMEOUT=15 # Timeout for detection summarization
SQ_VALIDATE_TIMEOUT=10 # Timeout for LLM validation
SQ_ANALYZE_TIMEOUT=8 # Timeout for LLM analysis
SQ_ANALYZE_TRACKING_TIMEOUT=10 # Timeout for LLM analysis with tracking
# General system timeouts
SQ_LLM_TIMEOUT=30 # Main LLM request timeout
SQ_CAPTURE_TIMEOUT=10 # RTSP capture timeout
SQ_TTS_TIMEOUT=5 # Text-to-speech timeout
# Vision model for image analysis
SQ_MODEL=llava:7b # Default vision model
SQ_VISION_MODELS=llava,moondream,bakllava,llava-llama3,minicpm-v
# Vision model confidence thresholds
SQ_VISION_ASSUME_PRESENT=0.5 # Default confidence when vision can't load
SQ_VISION_CONFIDENT_PRESENT=0.9 # Confidence for confident YES response
SQ_VISION_CONFIDENT_ABSENT=0.9 # Confidence for confident NO response
# Image preprocessing for vision models
SQ_IMAGE_PRESET=balanced # Preset: fast, balanced, quality, minimal
SQ_IMAGE_MAX_SIZE=512 # Max dimension in pixels
SQ_IMAGE_QUALITY=65 # JPEG quality 1-100
SQ_IMAGE_POSTERIZE=0 # 0=off, 8-256=reduce colors
SQ_IMAGE_GRAYSCALE=false # Convert to grayscale
SQ_FRAME_SCALE=0.5 # Frame scaling factor
# Guarder model for response filtering
SQ_GUARDER_MODEL=gemma:2b # Default guarder model
SQ_USE_GUARDER=true # Enable/disable guarder filter
SQ_ANALYSIS_MODEL=qwen2.5:3b # Model for response analysis
Guarder Filter Behavior:
--no-guarder flag# Object tracking parameters
SQ_TRACK_MIN_STABLE_FRAMES=3 # Frames before track is stable
SQ_TRACK_BUFFER=90 # Frames before deleting lost track
SQ_TRACK_ACTIVATION_THRESHOLD=0.25 # Min confidence for new tracks
SQ_TRACK_MATCHING_THRESHOLD=0.8 # IoU threshold for track matching
# Motion gating for tracking
SQ_MOTION_GATE_THRESHOLD=1000 # Min motion area to trigger detection
SQ_PERIODIC_INTERVAL=30 # Force detection every N frames
SQ_SKIP_INTERVAL=5 # Skip LLM check interval in other modes
SQ_CONSECUTIVE_NO_TARGET_SKIP=5 # Skip every Nth frame when no target detected
# Text-to-speech settings
SQ_TTS_ENGINE=auto # auto, pyttsx3, espeak, say, powershell
SQ_TTS_VOICE= # Voice name fragment
SQ_TTS_RATE=150 # Speech rate (words per minute)
# Speech-to-text settings
SQ_STT_PROVIDER=google # google, whisper_local, whisper_api
SQ_WHISPER_MODEL=small # tiny, base, small, medium, large
# Ollama configuration
SQ_OLLAMA_URL=http://localhost:11434 # Ollama server URL
# RTSP configuration
SQ_RTSP_TIMEOUT=10 # RTSP connection timeout
SQ_RTSP_RECONNECT_DELAY=5 # Delay between reconnection attempts
SQ_CAPTURE_FPS=1 # Default capture FPS
SQ_FAST_CAPTURE=true # Enable fast capture mode
# RAM disk configuration
SQ_RAMDISK_PATH=/dev/shm/streamware # RAM disk path for fast I/O
SQ_RAMDISK_ENABLED=true # Enable/disable RAM disk
SQ_RAMDISK_SIZE_MB=512 # RAM disk size in MB
# Logging settings
SQ_LOG_LEVEL=INFO # DEBUG, INFO, WARNING, ERROR
SQ_LOG_FILE= # Log file path (empty = stdout only)
SQ_LOG_FORMAT=yaml # Log format: text, yaml, json
SQ_TIMING_ENABLED=true # Enable timing logs
SQ_PERFORMANCE_ENABLED=true # Enable performance metrics
# Performance optimization
SQ_LLM_MIN_MOTION_PERCENT=30 # Minimum motion to trigger LLM
SQ_FAST_CAPTURE=true # Persistent RTSP connection
SQ_STREAM_MODE=track # Smart movement tracking
SQ_STREAM_FOCUS=person # Focus on person detection
# Resource management
SQ_MAX_CONCURRENT_LLM=2 # Maximum concurrent LLM requests
SQ_MAX_FRAME_QUEUE=10 # Maximum frame queue size
SQ_CLEANUP_INTERVAL=60 # Cleanup interval in seconds
# Custom prompt templates (override defaults)
SQ_PROMPT_STREAM_DIFF= # Custom stream diff prompt
SQ_PROMPT_STREAM_FOCUS= # Custom stream focus prompt
SQ_PROMPT_TRIGGER_CHECK= # Custom trigger check prompt
SQ_PROMPT_MOTION_REGION= # Custom motion region prompt
SQ_PROMPT_TRACKING_DETECT= # Custom tracking detection prompt
SQ_PROMPT_LIVE_NARRATOR_TRACK= # Custom live narrator track prompt
# Enable/disable features
SQ_USE_SMART_DETECTOR=true # Enable smart detector
SQ_USE_RESPONSE_FILTER=true # Enable response filter
SQ_USE_IMAGE_OPTIMIZATION=true # Enable image optimization
SQ_USE_ASYNC_LLM=true # Enable async LLM processing
SQ_USE_MOTION_GATING=true # Enable motion gating
# Optimized for speed and low resource usage
SQ_MODEL=moondream
SQ_IMAGE_PRESET=fast
SQ_YOLO_CONFIDENCE_THRESHOLD=0.2
SQ_LLM_MIN_MOTION_PERCENT=50
SQ_RAMDISK_SIZE_MB=256
SQ_TTS_ENGINE=espeak
# Optimized for maximum accuracy
SQ_MODEL=llava:13b
SQ_IMAGE_PRESET=quality
SQ_YOLO_CONFIDENCE_THRESHOLD=0.1
SQ_VISION_CONFIDENT_PRESENT=0.95
SQ_TRACK_MIN_STABLE_FRAMES=5
SQ_TRACK_BUFFER=120
# Optimized for systems with limited resources
SQ_MODEL=moondream
SQ_IMAGE_PRESET=minimal
SQ_FRAME_SCALE=0.25
SQ_RAMDISK_SIZE_MB=128
SQ_MAX_CONCURRENT_LLM=1
SQ_USE_ASYNC_LLM=false
Most configuration changes require restarting the application:
# Restart live narrator with new configuration
sq live narrator --url "rtsp://..." --restart
Some settings can be changed at runtime:
High false positive rate:
SQ_YOLO_CONFIDENCE_THRESHOLD=0.3
SQ_VISION_CONFIDENT_PRESENT=0.95
Missing detections:
SQ_YOLO_CONFIDENCE_THRESHOLD=0.1
SQ_VISION_CONFIDENT_PRESENT=0.8
Slow performance:
SQ_MODEL=moondream
SQ_IMAGE_PRESET=fast
SQ_LLM_MIN_MOTION_PERCENT=50
Memory issues:
SQ_RAMDISK_SIZE_MB=128
SQ_MAX_FRAME_QUEUE=5
SQ_MAX_CONCURRENT_LLM=1
# Check configuration
sq --check config
# Test specific settings
sq --check camera "rtsp://..."
sq --check tts
sq --check ollama
Last Updated: December 2024 (Post-Refactoring v2.0)