Google Cloud Free Tier & Sizing

Scout’s Google Cloud deployment script (./deploy/deploy-vm.sh) defaults to a Compute Engine e2-micro instance with a 30 GB standard persistent disk in us-central1-a. Here is what you need to know about Google Cloud’s Always Free tier and e2-micro resource sizing.

About the Google Cloud Free Tier

Google’s Always Free tier covers:

The hours of one non-preemptible e2-micro instance per month, per billing account (summed across every e2-micro in every project on that billing account), and only in:
- us-west1 (Oregon)
- us-central1 (Iowa)
- us-east1 (South Carolina)
30 GB-months of Standard Persistent Disk (pd-standard), shared across any standard disks on your billing account.
1 GB/month of outbound data transfer from North America (excluding destinations in China and Australia).

Important Billing Notes

One instance at a time: One e2-micro running all month uses the entire monthly hour allowance, so running a second e2-micro at the same time on the same billing account is billed.
What is billed at standard rates: Other regions or machine types (e2-small, e2-medium), pd-balanced or SSD persistent disks, outbound data transfer beyond 1 GB, and GPUs are billed at standard rates. Note that the external IPv4 address isn’t listed in the free tier.
Google Cloud Free Trial ($300 credit for 90 days): If you signed up through the Free Trial, upgrade to a Paid billing account before the trial ends. Otherwise, when the trial expires, the trial billing account closes and the VM is stopped—and then deleted after a 30-day grace period. Always Free tier usage continues after you upgrade to a Paid billing account.
Policy changes: Google can change Always Free limits with 30 days’ notice. Always check the current Google Cloud Free Program documentation and your billing reports/budget alerts before relying on a $0 bill.

Resource & Sizing Considerations (`e2-micro`)

Whether or not your e2-micro falls within the Always Free allowance, it is a shared-core burstable VM with specific hardware characteristics:

Resource	Default Configuration	Details
Machine Type	`e2-micro`	2 shared vCPUs (0.25 vCPU sustained baseline) + 1 GB physical RAM
Boot Disk	30 GB `pd-standard`	Standard Persistent Disk
Swap Memory	2 GB `/swapfile`	Auto-configured by `deploy-vm.sh` to guard against OOM spikes
Persistent Storage	`/opt/scout/data` → `/app/data`	Holds SQLite (`scout.db`) and uploaded/generated media

1. Memory Constraints (1 GB RAM + 2 GB Swap)

What fits comfortably: The FastAPI server, Caddy reverse proxy, SQLite database, and background worker run within ~300–450 MB of RAM under normal personal usage because LLM reasoning happens via remote cloud APIs (Gemini, Claude, OpenAI, etc.).
What will struggle: Compiling packages locally, building large Docker images on the VM, running local LLMs (such as Ollama), or running heavy headless browsers on the VM will exhaust 1 GB of RAM and thrash swap. That is why Scout pulls pre-built images from GHCR (ghcr.io/kcthota/scout) by default.

2. Burstable CPU (0.25 vCPU Baseline)

e2-micro provides 2 shared-core virtual CPUs operating on a credit-based burst model with a 0.25 vCPU sustained baseline.
Bursts of activity—handling webhooks, running a subagent turn, delivering reminders—are snappy, while sustained CPU-intensive loops will consume burst credits and throttle.

3. Ephemeral External IP Address

Standard Compute Engine instances receive an ephemeral public IPv4 address by default.
Restart impact: Rebooting Inside the OS keeps the IP, but stopping and starting the VM in the GCP Console reassigns a new external IP address.
Domain impact:
- If using the default <IP>.sslip.io domain, re-run ./deploy/deploy-vm.sh set-domain <NEW-IP>.sslip.io.
- If using a custom domain, update your DNS A record to the new IP address (or reserve a static external IP in Google Cloud VPC Network settings).

When to Upgrade Your Machine SKU

For heavier workloads or multiple concurrent background subagents, pass --machine-type during deployment or upgrade:

e2-small (2 vCPUs, 2 GB RAM): Recommended if you run frequent concurrent subagents or large knowledge document ingestion (~$7–$10/month).
Terminal window
```
./deploy/deploy-vm.sh --machine-type e2-small
```
e2-medium (2 vCPUs, 4 GB RAM): Recommended for heavy automation schedules and large MCP tool workloads (~$15–$25/month).
Terminal window
```
./deploy/deploy-vm.sh --machine-type e2-medium
```