Self-hosting

Run ADC on your own server

The self-hosted app is the same one that runs the official service. You need Docker Compose 2.24.4+, a hostname, ports 80/443 and persistent storage.

Deploy

Terminal

cp .env.example .env
openssl rand -hex 48   # ADC_AUTH_SECRET
openssl rand -hex 32   # ADC_DATABASE_PASSWORD
docker compose --env-file .env -f deploy/compose.yaml up -d --build

ADC_PUBLIC_URL: The public HTTPS origin.
ADC_AUTH_SECRET: A stable secret of 32+ characters. Back it up.
ADC_DATABASE_PASSWORD: PostgreSQL password (URL-safe).
ADC_SMTP_URL: Required when email verification is on.
ADC_TRUSTED_PROXIES: Only the proxies allowed to set client IP headers.
ADC_ANALYTICS_*: Leave unset to send no external analytics (the default).

Caddy obtains TLS certificates automatically. Set ADC_REGISTRATION_ENABLED=false once the intended accounts exist.

Verify

Terminal

curl -fsS https://devices.example.com/health
# {"ok":true,"schemaVersion":"0.1"}
docker compose --env-file .env -f deploy/compose.yaml logs --tail=100 control-plane

Then follow the Quickstart once: pair a device and complete one read-only call. A healthy /health alone does not prove sign-in, WebSocket upgrades and dispatch all work.

Operate

Probe /health from outside the host and alert on anything but 200.
Scrape the authenticated /metrics endpoint; never expose it publicly.
Alert on disk space and PostgreSQL availability. There is no automatic quota or retention yet.

Metric	Meaning
`adc_node_wake_connections`	Connected devices right now.
`adc_node_poll_total`	Device polls: dispatched, idle or denied.
`adc_invocation_total`	Calls by tool and outcome.

Back up and restore

Terminal

docker compose --env-file .env -f deploy/compose.yaml exec -T postgres \
  pg_dump -U adc -d adc -Fc > adc-backup.dump
docker compose --env-file .env -f deploy/compose.yaml exec -T postgres \
  pg_restore --list < adc-backup.dump

Keep the dump together with .env in encrypted storage, and test a restore into a separate database. Restoring can bring back credentials revoked after the backup, so rotate them after an incident.

Upgrade

Record the running version and take a verified backup.
Rebuild and recreate the control plane. Migrations run before HTTP starts.
Check /health, sign-in, device presence and one read-only call.
Keep the previous image and backup until you are satisfied; a schema rollback may need the backup.