YAML is your installation plan

Keep human-owned configuration separate from generated deployment files. Change the source YAMLs, validate them, then regenerate.

Which file controls what?

Actual values live in config/YAMLs. Samples are versioned under config/config_sample. The controller reads the installation directory by default.

File	What you configure
device.yaml	Device ID, environment, logging, and health reporting.
api.yaml	Branch ID/key, backend URL, ingest batches, retries, and buffer storage.
cameras.yaml	Enabled sources, camera IDs, FPS targets, model assignments, classes, zones, and table routing.
models.yaml	Weights, model ID/version, Ultralytics device selector, thresholds, and tracking choice.
boxmot_tracker.yaml	Tracking algorithm and its parameters.
rules.yaml	Camera/class/zone matching, UTC schedules, cooldowns, severity, and messages.
notifications.yaml	Log, webhook, and SMTP channels; rule and severity filters.
collection.yaml	Dataset sessions, output path, sampling, filters, and sample caps.
graph.yaml	Node templates, typed connections, optional paths, and queue profiles.

Source in, generated files out

Installation YAMLs, credentials, weights, datasets, and generated output stay local. Keep source code and samples in Git.

Project layout

config/
  config_sample/     # tracked templates
  YAMLs/             # your installation values
  controller/        # parser, validator, compiler, writer
  .generated/        # generated pipeline and Compose
models/              # local model assets
collected/           # dataset output

Catch mistakes before startup

The controller rejects unknown keys, invalid types, duplicate IDs, missing references, invalid class selections, and incompatible graph connections. Review file and field names in its error output.

Terminal

python -m config.controller --details
python -m config.controller --plan
python -m config.controller --generate
python -m config.controller --help