Argus

Settings

All settings are live — changes take effect immediately with no restart. Edit them via the Settings page in the UI, or via the API:

# Get a setting
curl -H "X-API-Key: argus_..." http://localhost:8100/api/settings/face.match_threshold

# Update a setting
curl -X PUT \
  -H "X-API-Key: argus_..." \
  -H "Content-Type: application/json" \
  -d '{"value": 0.6}' \
  http://localhost:8100/api/settings/face.match_threshold

Face recognition

Setting	Default	Description
`face.match_threshold`	0.5	Minimum cosine similarity to assign a match at all. Below this, the face is stored but left unidentified.
`face.auto_confirm_threshold`	0.80	Detections at or above this are confirmed automatically and bypass the review queue. Between `match_threshold` and this value, they land in the queue.
`face.auto_enroll_threshold`	0.92	Gate for automatic enrollment only. When Argus auto-confirms a match without human review, the embedding is added to the reference set only if face-detection quality clears this bar. Set to 0 to disable automatic enrollment. Human confirmations always enroll unconditionally.
`face.match_strategy`	`best_match`	`best_match` — score against every reference photo, take the closest. `average` — score against the mean embedding of all references.
`face.cluster_threshold`	0.5	Similarity threshold for the Suggested people clustering. Higher = stricter (more, smaller groups); lower = looser (fewer, larger groups). Adjustable live via the slider on the Suggested page.

System

Setting	Default	Description
`system.use_gpu`	true	When `true` and `CUDAExecutionProvider` is available, the face engine uses GPU. Setting to `false` forces CPU even on capable hardware. Setting to `true` on a system without GPU is rejected with a 400. The active provider is visible in `GET /api/health`.
`system.source_max_size`	1920	Maximum pixel dimension (width or height) for saved source images. Images larger than this are resized proportionally before storage. Detection always runs on the original full-resolution image before any resize occurs.
`system.ingest_jpeg_quality`	85	JPEG quality (1–95) used when saving source images.
`system.crop_jpeg_quality`	75	JPEG quality (1–95) used when saving face and object crops. Crops are taken from the original full-resolution image before source resize, preserving quality.
`system.crop_padding`	0.2	Fractional padding added around each detection bbox before cropping (e.g. 0.2 = 20% of bbox width/height on each side). Clamped to image bounds.
`system.url_fetch_timeout_seconds`	10	Timeout in seconds for fetching images via `image_url`. Private/local IP ranges are intentionally allowed — Argus is designed to pull from local cameras and services.
`system.url_fetch_max_size_mb`	25	Maximum size in MB for images fetched via `image_url`.

Object detection

Setting	Description
Object classes	For standard YOLO models, a filter list of the 80 COCO class names to detect. Leave empty to detect all 80. Ignored for YOLO-World models.
YOLO-World vocabulary	For YOLO-World models, the list of plain-language descriptions to detect. One per line. Changes take effect on the next detection — no restart needed. See YOLO-World in the Concepts guide.