Video Analysis Settings
Video Analysis controls how Aegis AI processes recorded clips using the active VLM. These settings determine when analysis runs, how often frames are extracted, and what the model looks for.
Access via ⚙️ Settings → System → AI Engine → Video Analysis, or in the Settings modal.
Settings Reference
Enable Video Analysis
Master toggle. When disabled, clips are recorded but not analyzed — no descriptions are generated.
Default Analysis Prompt
The text prompt sent to the VLM with each extracted frame. Controls what the model focuses on.
Default: "What's in this image? Focus on security-relevant details..."
You can customize this to focus on specific things:
"Describe any people, vehicles, or animals visible. Note any unusual activity.""Is there a package on the porch? Describe its size and location."
Frame Rate
How many frames per second are extracted from clips during analysis.
| Value | Behavior |
|---|---|
| 0.1 fps | 1 frame every 10 seconds — minimal resource usage |
| 1.0 fps | 1 frame per second — balanced (default) |
| 5.0 fps | 5 frames per second — detailed but resource-intensive |
The slider ranges from 0.1 to 5.0 fps. The actual rate adapts automatically based on how fast the VLM can process frames — if inference is slower than the target rate, frames are skipped.
Temperature
Controls randomness in the VLM's analysis output.
| Value | Behavior |
|---|---|
| 0.0 | Most consistent — same scene always gets the same description |
| 0.1 | Slight variation (default) |
| 0.5 | Balanced |
| 1.0 | Most creative — may notice different details each time |
For security use, keep this low (0.0–0.2) for consistent, reliable descriptions.
Max Analysis Length
Maximum number of characters per frame analysis. Range: 100–2000. Default is usually around 500.
Shorter limits produce faster responses. Longer limits produce more detailed descriptions.
Auto-start Analysis
When enabled, analysis begins automatically when you open a clip in the player. When disabled, you must manually start analysis from the clip playback view.
Test Analysis
Click Run Test to verify the VLM connection and analysis pipeline. The test:
- Sends a test frame to the active VLM
- Returns a pass/fail result with the model's response
- Shows detailed test output in an expandable section
The test button is disabled when:
- Video Analysis is toggled off
- The LLaMA server is offline (for local models)
LLaMA Server Status
A status indicator at the top shows whether the local llama-server is:
- Online (green) — ready to analyze
- Offline (red) — needs to be started from the VLM settings