LLM Notifier for Scrypted

Enhance security camera notifications with AI-generated descriptions using vision-capable LLMs.

Setup

1. Install LLM Plugin - Required dependency for ChatCompletion providers
2. Configure LLM Provider - Select OpenAI, Claude, or local LLM in LLM Notifier settings
3. Enable on Devices - Extensions → toggle on your notification devices
4. Test - Trigger a detection to see enhanced notifications

Settings

• LLM Providers: Select one or more (rotates between them for load balancing)
• Notification Style: Customize how detections are described (see below)
• Snapshot Mode: Cropped (fast), Full (context), or Both (accurate)

Notification Style Prompts

v12 (Latest)

Focus: License plate handling and person identification

STYLE PREFERENCES:

Title: Include person names ONLY when provided in metadata, otherwise use generic terms
- When name known: "Richard at front door"
- When name unknown: "Person at front door"
- For vehicles: Include actual license plate only if clearly visible

Subtitle: Category marker
- Format: "[Type] • [Area]"
- Examples: "Person • Indoor", "Vehicle • Street"

Body: Focus on actions and key visual details
- Describe what's happening in the scene
- Include relevant clothing, objects, or movements
- Include license plate details only if actually visible
- Examples:
  "Walking toward garage while checking phone and carrying a shopping bag"
  "White sedan pulling slowly into space with headlights illuminating wall"
  "Tall figure in blue jacket with package approaching and ringing doorbell"

Common locations: driveway, street, kitchen, living room, front door, yard, garage

Avoid generic phrases like "motion detected" or "person detected"

v11

Focus: Person identification and generic terms

STYLE PREFERENCES:

Title: Include person names ONLY when provided in metadata, otherwise use generic terms
- When name known: "Richard at front door"
- When name unknown: "Person at front door"

Subtitle: Category marker
- Format: "[Type] • [Area]"
- Examples: "Person • Indoor", "Vehicle • Street"

Body: Focus on actions and key visual details
- Describe what's happening in the scene
- Include relevant clothing, objects, or movements
- Examples:
  "Walking toward garage while checking phone and carrying a shopping bag"
  "White sedan pulling slowly into space with headlights illuminating wall"
  "Tall figure in blue jacket with package approaching and ringing doorbell"

Common locations: driveway, street, kitchen, living room, front door, yard, garage

Avoid generic phrases like "motion detected" or "person detected"

Features

• Face Recognition: Preserves names from "Maybe: Richard" metadata
• Vehicle Details: Make, model, color, and real license plates when visible
• Animal Breeds: Identifies breeds when clear
• Character Limits: Optimized for mobile notifications (32/32/80 chars)
• Platform Support: Works with any Notifier device (Pushover, Home Assistant, etc.)

Troubleshooting

• No enhancement: Check LLM provider is vision-capable
• Timeouts: Reduce LLM timeout or use single provider
• Wrong names: Ensure "Include Original Message" is enabled