The diagram is the source now — wire resources on a canvas, and whiparc emits Terraform and Ansible you can read, review in a PR, and run.
Keep scrolling — the three states below are the entire product. Nothing here is a mock of a feature we haven't built.
Drop resources, wire dependencies. The canvas is typed — it refuses edges that wouldn't plan.
Every edit re-emits HCL and YAML. Ordinary files — diff them, review them, hand-edit them.
Plan and apply from a sandbox agent with your credentials, streaming the real log.
You'd find these out in twenty minutes anyway. Better you read them here and decide before you invest an afternoon. Keep scrolling — they come one at a time.
Roughly forty AWS resources are modelled properly — the ones you'd actually wire in a three-tier stack. GCP and Azure cover compute, network and storage, and that's the honest end of the list.
You can't point whiparc at a live account and get a canvas back. Everything starts from a blank canvas or a template, which makes it a tool for new stacks and rebuilds rather than for documenting what you already run.
Projects, members and access requests work. In-canvas comments and per-environment approvals don't exist — we'd rather the emitted code go through the review flow your team already trusts.
Linux and macOS agents are steady through long applies. On Windows, a thirty-minute apply sometimes loses the log stream and you have to re-attach to see the tail — the apply itself keeps running.
So much gets built and just sits on localhost — the deployment step is where most people stall. We're two people who spent years deep in other teams' Ansible playbooks and Terraform modules, and whiparc is the tool we wanted then: a single canvas to learn how the cloud actually fits together, and to ship for real, with real files and no layer in between that you can't read.
VPC, two private subnets, ALB, autoscaled app tier, RDS. The one most people start from.
One box, k3s, ingress and a playbook that installs it. Cheap staging that behaves like production.
S3, CloudFront, ACM cert and a DNS record. Twelve lines of HCL you'd rather not write again.
Every template is a normal project once you fork it. Rename things, delete the RDS node, swap the region — it re-emits and you own the output.
Self-hosting is the whole product, permanently, with no seat cap and no feature held back. The hosted plan exists because some teams don't want to run another service — that's the only difference.
One Docker command. Your machines, your credentials, your data — we never see any of it.
Same product, on our infrastructure, with managed agents and backups. Cancel and export everything as plain files.
No credit card to start · no usage metering on the editor · cloud bills stay between you and your provider