1.Infrastructure as code

A singleTerraformroot orchestrates four components: credentials,cloud-initpayloads, VM launch, and user provisioning. Your hub lives entirely in your cloud account. We publish every module on the Terraform Registry, with provider-specific variants for your preferred cloud.

module "flexpair" {
  source        = "Flexpair/flexpair/<your-cloud>"
  region        = "eu-central-1"
  desktop_count = 5
  admin_email   = "you@example.com"
}

2. Persistent, not ephemeral, and as many as you want

Desktop VMs keep running. Your Jupyter kernels, long-running trainings, SSH tunnels, even your terminal scrollback: all still there tomorrow, and next week. When a teammate opens the session, nothing has been reset. Spin up one desktop per project, one per engineer, or one per training run; the hub scales to whatever your team needs.

The continuous uptime of a single hub is bounded by the TLS certificate it was provisioned with. We fetch 90-day Let's Encrypt wildcard certificates via DNSimple, which auto-renews 30 days before expiry, so a hub is deployed with somewherebetween 30 and 90 days of TLS validity, depending on where in the renewal cycle you run terraform apply. Before the current certificate expires, the pipeline rolls a fresh hub with the renewed certificate; persistent data volumes carry the session state across.

3. Shared cursor per desktop, one voice channel across the hub

Built on Apache Guacamole with a shared-cursor protocol. On any given desktop, everyone who is on it drives the same mouse and keyboard. Mumble runs as an always-on voice channelacross the whole hub. No “unmute to join”, no video noise. Just step into a desktop and start talking.

4. Your data, your perimeter

No data ever transits through Flexpair infrastructure. Authentication goes through your IdP. Traffic is wildcard-TLS end-to-end. Because you own the VMs, compliance scopes (GDPR, SOC 2, ISO 27001) stay within boundaries you already audited.

5. Deploy time you can count on

Across the last 17 production applies on Terraform 1.14, a full hub deployment took on average 967 s (≈ 16 minutes) with a standard deviation of 64 s (≈ 1 minute). The 95 % prediction interval for your nextterraform apply is 14.0 to 18.2 minutes. That is the entire pipeline: credentials, cloud-init rendering, VM launch on AWS, and Guacamole user provisioning, end to end.

Historical context: earlier Terraform versions (1.5.7, 1.9.8) used to average 24 to 28 minutes with swings of several minutes per run. The 1.13 and 1.14 line cut that in half and, more importantly, made the outcome predictable enough that a deploy window fits in a single coffee break.