Requirements

FlowDesk is installed by running three containers (API, Dashboard, Tracker) on a Kubernetes cluster. Which cloud provider or deployment tooling you use is your organisation’s choice; as long as the requirements described here are met, the application behaves as expected.

Architecture

browser ──HTTPS──▶ [your ingress] ──┬──▶ flowdesk-web      :3000 ──┐
                                    ├──▶ flowdesk-tracker  :3001 ──┤ (HTTP, server side)
                                    └──▶ flowdesk-api      :8080 ◀─┘
                                                │
                      PostgreSQL or SQL Server · Redis · RabbitMQ or Kafka
                                                │
                    outbound ──▶ licence server · Keycloak (if used) · SMTP/IMAP (if used)

All three containers listen on plain HTTP and expect no TLS: the ingress in front of them terminates the certificate. Logs are written to stdout.

Kubernetes cluster

• The shftco/flowdesk-kubernetes chart uses apps/v1 Deployment and networking.k8s.io/v1 Ingress resources, which in practice means it works with Kubernetes 1.19 and later. SHFT documents no separate lower bound; any compatible cluster will do, k3s included.
• ingress-nginx: the Ingress resource the chart renders expects ingressClassName: nginx.
• cert-manager: manages TLS certificates through a ClusterIssuer (Let’s Encrypt, for example). If cert-manager is not installed, or you do not name an issuer, you have to manage TLS yourself.
• A default StorageClass: needed to provision persistent disks for the database and for file attachments (on k3s, local-path is enough).

Database

PostgreSQL 14+ or SQL Server 2019+: you pick which one at install time with Database:Provider (see Configuration). Both run from the same image; changing later means migrating data, so decide up front. The chart ships with an in-cluster PostgreSQL of demo/single-replica quality (postgres.enabled: true); to use a managed database in production (RDS, Cloud SQL, Azure Database and so on) you turn that off and supply your own connection string.

Redis

Redis 7+. Used for session/refresh tokens and general caching. The chart includes an in-cluster Redis of demo/single-replica quality (redis.enabled: true); you can turn it off and point at your own. Do not disable Redis in production.

Message queue: RabbitMQ or Kafka

Background work (notifications, automation, inbound e-mail processing) runs over a message queue. The choice is made with Messaging:Provider:

• RabbitMQ (default): the chart includes an in-cluster RabbitMQ 3.x of demo/single-replica quality, or you can point at your own.
• Kafka: you can point at your own Kafka cluster, or enable the chart’s single-node in-cluster Kafka intended for demos (messaging.kafka.inCluster.enabled). In a Production environment a non-local broker cannot run over an unencrypted protocol (plaintext); the API refuses to start.

Authentication

The FlowDesk API supports e-mail/password, LDAP and Keycloak (OIDC) login providers (Auth:Providers). If you are going to use Keycloak you need your organisation’s own Keycloak server (26+ recommended): realm, client and user management stay entirely with you. For details see Configuration.

E-mail (optional but recommended)

An SMTP account for outgoing notifications (welcome e-mail, one-time passwords); if you also want requests opened automatically from incoming mail, a mailbox with IMAP access. These settings are entered from the dashboard and need no environment variable; see E-mail settings.

Resource expectations

The default requests/limits values in the chart’s values.yaml are intended for a single-replica, demo/small-scale installation; tune them in values.yaml for your production load.

Component	CPU (request / limit)	Memory (request / limit)
API	250m / 1	256Mi / 1Gi
Dashboard (web)	100m / 500m	128Mi / 512Mi
Tracker	100m / 500m	128Mi / 512Mi
PostgreSQL (if in-cluster)	250m / 1	256Mi / 1Gi
Redis (if in-cluster)	50m / 250m	64Mi / 256Mi
RabbitMQ (if in-cluster)	100m / 500m	256Mi / 512Mi

Running everything in-cluster, total requests come to roughly 0.85 vCPU / ~1.1Gi memory and total limits to roughly 3.75 vCPU / ~3.75Gi memory.

Persistent disk (PVC) needs:

Purpose	Default size	Impact if lost
API: file attachments (/app/attachments)	5Gi (grows over time)	Attachments and archived inbound e-mail are lost
API: licence state (/var/lib/flowdesk)	1Gi	Not critical; the installation re-registers with a new identity
PostgreSQL (if in-cluster)	5Gi	All data is lost

Network requirements

Inbound (through the ingress): three separate hostnames, for the dashboard, the tracker and the API. Raise the request body size limit on your proxy (25 MB recommended): the default 1 MB limit truncates file attachments. The optional Knowledge Base adds three hostnames and a database of its own; see Knowledge Base.

Outbound (from the application):

Target	What for	Required
ghcr.io	Pulling images (at deploy time only)	Yes
license-api.flowdesk.com.tr	Licence check, HTTPS, roughly every 12 hours	Yes (it also runs unlicensed, see Configuration)
Your Keycloak server	Authentication	If Keycloak is used
Your SMTP/IMAP server	Outgoing/incoming e-mail	If e-mail is used

There is no connection directed from SHFT into your organisation’s systems; traffic is one-way and outbound from your organisation.

Decide before you start

• The image tags (versions) you will run; see Release Notes
• Hostnames for the dashboard, the API and the tracker
• Database: PostgreSQL or SQL Server
• Authentication provider: EmailPassword, LDAP or Keycloak
• Message queue: RabbitMQ or Kafka
• Whether you will run the optional Knowledge Base: it needs its own hostnames, database and decisions

When you are ready, move on to Installing with Helm.