Hardening checklist
A step-by-step checklist to secure an Orchestrator deployment, its Robots and its users before you go live.
Work through this list after you install the Orchestrator and before you run production automations. Each item links to the page that explains it. Items are grouped in the order you would normally do them.
Before the first start#
- Put the deployment directory on a host only administrators can reach, and set
chmod 600 .env. - Generate every secret with
openssl rand -base64 32(or-base64 24for the database and Redis passwords). Never reuse a value from an example or another environment. - Escrow
VELOPHEX_KEK,VELOPHEX_KEK_IDandVELOPHEX_TOKEN_PEPPERin your organization's secret escrow, with two-person access and a written recovery procedure. See Secrets and encryption. - Set
VELOPHEX_PUBLIC_BASE_URLto anhttpsorigin andVELOPHEX_SITE_ADDRESSto its host name. - Choose how TLS is issued: Let's Encrypt for a public name,
VELOPHEX_TLS=tls internalwith Caddy's root trusted on every machine, or your own certificate. - Keep the
velophex_caddy_datavolume persistent so certificates survive restarts. - Check the image digests after
docker compose -f docker-compose.core.yml pull, and verify release files againstSHA256SUMS.
Network#
- Allow inbound TCP 443 only (plus 80 for Let's Encrypt). Block 8080, 5432 and 6379 from outside the host. See Network and firewall.
- If you use your own load balancer, confirm all eight settings on the network page, especially the trusted proxy network and WebSocket forwarding.
- Give Robot machines an outbound route to the Orchestrator on 443, directly or through a proxy set on the Robot. Open nothing inbound on them.
First sign-in#
- Sign in with the bootstrap administrator and change its password immediately (your profile, Security). Set up two-step sign-in or a passkey on the same page.
- Remove the three
VELOPHEX_BOOTSTRAP_ADMIN_*lines from.envand rundocker compose -f docker-compose.core.yml up -d api. - Set
VELOPHEX_EMAIL_MODE=Smtpwith a working relay, so invitations and password resets go by mail rather than to the log. - Review the
securitysettings group: failed sign-ins before lockout, lockout duration, password length and complexity, password expiry, session timeout, andrequireMfato require a second factor for local accounts. See Settings. - If you use an identity provider, add it under Platform admin, Identity providers, and consider SCIM provisioning so leavers lose access automatically. See Users and roles.
- In the tenant's services, turn on only the services you use.
Access control#
- Give the Administrator role to as few people as possible. Use Workspace Administrator for team leads.
- Put automation teams in their own workspaces and assign roles at workspace scope.
- Assign roles to groups rather than to individuals, so leavers are removed in one place.
- Create personal access tokens with the smallest permission set and a short expiry. Give CI publishing tokens only
packages.viewandpackages.create. - Review members, roles and tokens on a schedule. See RBAC and permissions.
Robots#
- Turn on Bind to a single host for each machine, and sysprep VM images before cloning.
- Deliver enrollment keys through your deployment tool as MSI properties. Never put them in scheduled tasks or scripts in source control.
- Assign each machine only the workspaces whose jobs it should run.
- Keep the service on LocalSystem (the supported account) and give each job a dedicated run-as account with only the rights the automation needs.
- Decide your package signing policy. If you require signatures, pin signers first, then set
REQUIRESIGNED=1orconfigure --require-signed true. See Package signing and supply chain. - Install only activity packages you trust. Their declared capabilities are a consent signal, not an isolation boundary.
- Revoke a machine as soon as it is retired or suspected compromised.
Secrets#
- Store every password the automations use as a Credential asset, never in workflow files,
project.jsonor queue item payloads. - Plan KEK rotation and tenant data key rotation, and record who can run them.
- Use encrypted disks or storage-level encryption for the database, blob store and backups.
Operations#
- Schedule nightly cold backups and copy them off the host. Include the escrowed
.envin your recovery plan. See Backup and restore. - Run a restore drill before go-live and once per quarter.
- Alert on
"Level":"Error"lines from the background workers and on/health/readynot returning 200. See Monitoring and health. - Set an audit window that matches your policy (90 to 2555 days). See Retention.
- Subscribe a webhook to
com.velophex.job.dead_letteredandcom.velophex.robot.offlineso failures reach your on-call. See Webhooks.
Verify#
When you have finished:
- From outside the host,
curl -I https://<your host>should showstrict-transport-securityin the response headers. nc -zv <your host> 8080from another machine should fail to connect..envshould contain noVELOPHEX_BOOTSTRAP_ADMIN_*lines.- On a Robot,
VeloPhex.Robot.Service.exe statusshould showconnectivity: Connected.
Next steps#
- Keep up to date: Upgrades and release channels.
- Report a problem you find: Responsible disclosure.
Before the first start
Network
First sign-in
Access control
Robots
Secrets
Operations
Verify
Next steps