You ask, it prepares
"Create a WordPress server with a public IP for me", "open port 443 on my web servers", "add two workers to my Kubernetes cluster": the assistant understands the request, looks up your resources and prepares the operation. It also answers your questions about the services, how to use them and their limits.
What it can do
It creates and manages your VMs (including with WordPress or Nextcloud installed on first boot), their firewalls and public IPs, volumes, security groups, load balancers and their certificates, Kubernetes clusters, S3 buckets and their keys, IPsec and WireGuard VPNs, autoscaling instance groups, snapshots and backups.
You stay in control
The assistant does nothing on its own: each operation is displayed as a card, with a summary calculated by the portal, and only your click on "Confirm" runs it. It acts with your identity and the permissions of your role: a read-only account cannot create anything, even by asking the assistant.
It never deletes anything and never logs into your VMs. Secrets, such as an S3 access key, never go through the AI: the portal shows them to you directly, only once.
Your data
The assistant relies on a cutting-edge AI model, Claude, from Anthropic. Your messages and the information about your resources that it looks up are sent to this provider to produce the answer; the content of your VMs and their boot data never are. Using it is optional: everything can still be done from the portal, the API or Terraform.
AI-assisted operations, under human control
On PYSOFT's side, a second AI agent monitors the platform in read-only mode: it analyzes metrics and logs, reports anomalies and proposes actions, which only an engineer can decide on. It has no access to the servers or to your VMs.