Frozen document. Dates and status refer to its original source. Private resources and unreviewed images are unavailable.
Free-tier hosting assessment
Historical free-tier comparison. ADR 0010 selects local certification followed by hosted UI and cloud VPS; it assumes no free-tier entitlement. The proposal below remains research, not the current deployment plan.
Status: researched recommendation, 2026-09-23; no provider selected or provisioned. Assess free-tier capacity and single-provider hosting for Atlas records the answer. The owner reports an existing AWS account created sometime in 2025; its exact date, plan, credits and status are unknown.
Answer
AWS can host every backend component that we operate, but ordinary free compute allowances are not enough for the complete Atlas stack. ECS does not create a separate pool of free machines. Traditional ECS on EC2 adds no orchestration fee, while the EC2 instances, storage and related resources remain billable. Fargate meters requested compute and memory. ECS Managed Instances is a different option with an additional management charge. ECS pricing.
Temporary AWS credits may fund appropriately sized infrastructure. That is different from an indefinitely free deployment. Furthermore, Contentful and the accepted MuleSoft CloudHub service remain separate providers/accounts; their underlying use of AWS does not bring their subscriptions into the owner's AWS free tier.
The existing 2025 account
- If created before July 15, 2025, its legacy 12-month introductory window has expired by this research date.
- If created on or after July 15, 2025, its six-month Free account plan window has also elapsed. Paid-plan credits may remain if their exact 12-month expiry has not passed and credits are still available; inspect Billing rather than assuming a balance.
- New-customer signup credits are $100 initially plus up to $100 earned through specified activities, not an automatic $200 balance. AWS says existing and former customers are ineligible for a new Free plan or new Free Tier credits, so another email/account does not reset eligibility.
- Lack of usage does not restart those account-age windows. Separate service offers can have different eligibility.
Sources: AWS Free Tier FAQ, EC2 account-date rules.
Compute options
| Option | What is actually free or credit-funded | Fit for Atlas |
|---|
| ECS on EC2 | Traditional EC2 launch type has no extra ECS orchestration fee | Choose and pay for adequate underlying capacity, possibly offset by valid credits |
| EC2 T4g.small trial | Up to 750 aggregate instance-hours/month through December 31, 2026, subject to trial terms; 2 GiB RAM | Too small for the full concurrent stack; ARM compatibility and ancillary charges still matter |
| New-plan eligible EC2 types | Includes larger 4 GiB and 8 GiB types, but usage consumes finite credits | Larger or multiple instances may run a short experiment; the label does not make them free forever or renew this owner's eligibility |
| ECS Fargate | Metered vCPU/memory with possible eligible credit offsets | Feasible compute, not a recurring free allocation for this stack |
| OCI Ampere A1 Always Free | Current documentation describes 2 OCPUs / 12 GB and 200 GB aggregate block storage allowance | Best continuing free compute candidate found, subject to quota, regional capacity and a successful workload test |
| Google Cloud recurring e2-micro offer | Standard 1 GB VM within eligible-region/time limits | Too small; introductory credits are a separate temporary option |
Sources: T4g trial and specifications, EC2 trial FAQ, Fargate pricing, OCI current limits, Google free program, Google machine sizes.
The T4g trial is not 750 hours for each instance or each region. Storage, public IPv4, logs, transfer and surplus CPU credits can still generate charges. Do not use swap or frequent container crashes as evidence that a 2 GiB instance is adequate.
Why the stack needs more memory
ClamAV documents 3 GiB minimum and 4 GiB preferred for its container, including signature reload behavior. Camunda, Keycloak, PostgreSQL and the backend/workers need additional memory. Our 16 GiB proposal is a starting budget, not a measured minimum; roughly 10–14 GiB is a sizing hypothesis to test for the lightweight stack. A 12 GB OCI VM is therefore plausible but tight, particularly during scanning updates and engine startup. ClamAV requirements, Camunda lightweight configuration.
OCI's current operational documentation differs from older 4 OCPU / 24 GB marketing. Use the lower current documented allowance until the specific tenancy proves otherwise. Home-region capacity is not guaranteed and idle instances can be reclaimed. Verify supported ARM image manifests, explicit memory limits, durable data, restore procedures and an actual combined workload before selecting it.
Recommended next choice
For an AWS-focused short prototype, use the existing account only after checking its remaining credits and service access. Prefer a suitably sized EC2 host with Docker Compose for this single-host evaluation; use ECS on EC2 if learning ECS is itself a requirement. Avoid adding ALB, NAT Gateway, RDS or EKS by default merely to run the first synthetic demo. Their benefits and charges need a separate reason. Keep the same secure HTTPS entry point, private databases/scanner and narrow service credentials described in the cloud proposal.
For continuing zero-charge infrastructure, evaluate OCI Always Free capacity first rather than attempting to fit the full stack on T4g.small. This is a proposed alternative, not approval to change providers. Preserve CloudHub and Contentful as required external services, and explicitly resolve CloudHub's post-trial funding. Without changing that vendor choice, the entire portal cannot be promised at zero recurring cost.
Detailed evidence: AWS account eligibility and costs and capacity/alternative research. No AWS/OCI account was accessed or created, no credentials were read, and no infrastructure was provisioned during this research.