Deploy on GCP

GCP is the recommended cloud for PandaStack. n2d-highcpu instances expose nested KVM with zero quota requests, and boot times are 175–205 ms (the same as bare metal).

Prereqs

• gcloud CLI logged in (gcloud auth login)
• A billing account
• The compute and iam APIs enabled

One-command deploy

git clone https://github.com/pandastack-io/pandastack
cd pandastack
./deploy/deploy-gcp.sh

The script will:

1. Create a project (or reuse one set in GCP_PROJECT)
2. Create a VPC + firewall rules
3. Provision a n2d-highcpu-32 VM in us-central1-a (Ubuntu 24.04)
4. Run cloud-init to install Firecracker, mount the XFS-reflink loopback, lay down pandastack-agent, pandastack-api, pandastack-dashboard, and Caddy
5. Bake the first-party templates: base, code-interpreter, agent, browser, and postgres-16
6. Issue Let's Encrypt certs for *.pandastack.ai or a domain you own

Total time: ~12 minutes.

Costs

For dev/staging, run --machine-type n2d-highcpu-8 (~$135/month, supports ~30 concurrent sandboxes).

Configuration

GCP_REGION=europe-west4 GCP_MACHINE=n2d-highcpu-64 PANDASTACK_DOMAIN=pandastack.acme.com ./deploy/deploy-gcp.sh

Clients should then use the API endpoint you expose:

export PANDASTACK_API=https://api.pandastack.ai
export PANDASTACK_API_KEY=pds_abc123def456...

Tearing down

./deploy/deploy-gcp.sh --destroy

This deletes the VM and zone — cert state in CF is preserved.