Cloud Spectra Gateway -- Quick Start v1.0.0

Deploy Cloud Spectra Gateway into your own AWS account in under 15 minutes.

1. Overview

Cloud Spectra Gateway is an AWS networking appliance that deploys entirely into your own AWS account. It replaces metered AWS networking and LLM-API spend with a fixed EC2 cost -- your cloud, off the meter. All traffic and configuration stay inside your account: there is no vendor control plane and no data leaves your boundary.

You subscribe on the AWS Marketplace, run one terraform apply, and within minutes you have a running gateway with a management dashboard. Outbound NAT works immediately; you turn on additional features from the dashboard as you need them.

Cloud Spectra Gateway replaces the AWS services you are currently metered on with in-appliance equivalents:

What AWS metersCloud Spectra Gateway equivalentTier
NAT Gateway (hourly + per-GB)Source NAT (sNAT) on the applianceStandard
Inbound forwardingDestination NAT / port forwarding (dNAT)Standard
Network Load Balancer (L4)In-appliance Linux IPVS L4 load balancingStandard
TLS terminationHAProxy + ACM certificate on port 443Standard
Inline inspection / IDS-IPSSuricata IDS/IPS (ET Open rules plus community threat feeds), nftables firewall, TLS inspection, domain and URL-category filtering, ClamAV antivirus, L7 App-ID, and SIEM exportAdvanced
Forward web proxySquid proxy with caching and filteringStandard
LLM API spendOpenAI-compatible AI Gateway with an exact-match response cache and a semantic cache, plus an in-dashboard AI Assistant backed by Amazon BedrockAdvanced
Always-on, oversized stateful box (database, search, analytics, dashboard)ElasticApps: one stateful app, vertically right-sized and on SpotStandard

Cloud Spectra ships as one product, and every deployment gets every feature: the compute and Kubernetes worker stack (CPU and GPU), the ElasticApps singleton, the networking data plane, inline inspection and enforcement, and the governed AI LLM gateway. The Standard and Advanced badges you will see throughout these docs are usage meters -- they say which line of your invoice an hour running that feature lands on, not whether you may run it. For the full breakdown of every feature and how it works, see the User Guide; for how the appliance is assembled internally, see the Architecture guide.

ElasticApps Standard

ElasticApps is a configuration choice for a different job: making one stateful open-source application cost-elastic without clustering or a front-end load balancer. You pick a single app -- Grafana, ClickHouse, OpenSearch, PostgreSQL, Neo4j, Trino, Prometheus and more -- and the Cloud Spectra control plane runs it on a single instance that is vertically right-sized: it moves the app onto a bigger instance when it is busy and a smaller one when it is idle, and it can run that instance on Spot for a deeper discount. A sticky EBS data volume follows the app across every resize and replacement, and a stable Elastic IP and DNS name keep its address unchanged -- so the instance can move to the right size, and across Availability Zones, while the data and the endpoint stay put. Because the software fee is charged per vCPU, scaling the box down directly shrinks your bill.

flowchart LR
    APP["One stateful app
(Grafana / ClickHouse /
Postgres / Trino / ...)"] --> Q{"App load?"} Q -->|"busy"| UP["Move to a
bigger instance"] Q -->|"idle"| DOWN["Move to a smaller
instance"] UP --> KEEP["Sticky EBS keeps the data
Elastic IP + DNS keep the address"] DOWN --> KEEP KEEP --> APP style APP fill:#dbeafe,stroke:#3b82f6,color:#1e3a8a style UP fill:#d1fae5,stroke:#10b981,color:#065f46 style DOWN fill:#d1fae5,stroke:#10b981,color:#065f46 style KEEP fill:#fef3c7,stroke:#f59e0b,color:#92400e
Deploying an elastic app? You subscribe and apply exactly as below (steps 2--5), then turn ElasticApps on and select the application and its scaling policy from the dashboard. Unlike the per-AZ network topology shown below, ElasticApps runs as a single multi-AZ instance -- no Gateway Load Balancer and no horizontal scale-out. See ElasticApps in the User Guide for the full configuration reference and Pricing for the fee.

CPU and GPU worker nodes

The CPU and GPU worker stacks -- both part of the Standard tier -- are worker-node AMIs rather than appliances. They exist to give a Kubernetes cluster ready-to-join nodes without you having to bake and maintain the image yourself, so there are no runtime settings to configure in the dashboard once a node is launched -- you launch the AMI and the node joins your cluster.

  • The CPU worker stack -- a Kubernetes worker AMI with the containerd runtime, kubeadm/kubelet ready to join a cluster, and the Calico CNI pre-installed.
  • The GPU worker stack -- the same worker AMI plus the NVIDIA driver, the CUDA toolkit, the vLLM runtime, and the NVIDIA device plugin pre-installed, for GPU workloads such as model serving.

Both are included in every deployment, and their hours bill on the Standard meter. The sticky EBS data volume, which re-homes a persistent volume onto a replacement instance without reformatting, is available to them like everything else.

Deploying worker nodes? Worker nodes have no module or dashboard of their own -- launch the AMI the way you launch any other worker node. The rest of this walkthrough follows the gateway path.

Deployment flow at a glance

graph TD
    MP["AWS Marketplace
subscribe to the product"] --> CF["terraform apply
vpc_id empty = new VPC, set = existing"] CF --> ASG["Per-AZ Auto Scaling Group
one ASG per Availability Zone"] ASG --> GWLB["Gateway Load Balancer (GENEVE)
horizontal scale"] GWLB --> EIP["Elastic IP
stable endpoint"] EIP --> DASH["HTTPS dashboard
port 9443 (nginx, self-signed)"] DASH --> DONE["Login & configure"] style MP fill:#dbeafe,stroke:#3b82f6,color:#1e3a8a style CF fill:#dbeafe,stroke:#3b82f6,color:#1e3a8a style DONE fill:#d1fae5,stroke:#10b981,color:#065f46

2. Prerequisites

To complete this quick start you need:

  • An AWS account where you can subscribe on the Marketplace and create resources.
  • A chosen AWS Region to deploy into (any commercial region works).
  • Permissions to create the resources the module declares: EC2 instances, VPC networking, IAM roles, Auto Scaling groups, and an Elastic IP.
  • No EC2 key pair needed. The stack creates its own key pair and stores the private key in SSM Parameter Store for you; the stack Outputs include the command to download it (see section 5).
  • Optional: an AWS Certificate Manager (ACM) certificate if you want TLS termination on port 443, in front of your own services, to serve a trusted certificate. This does not change the management dashboard's certificate -- in v1 the dashboard always serves a per-instance self-signed certificate.
Enable most features -- deploy the operational IAM role. The base module is intentionally minimal: the gateway boots and outbound NAT works from it, but most other features -- the full per-AZ NAT data plane for your private subnets, Gateway Load Balancer, EventBridge, vertical/horizontal scaling, the AI Gateway, teardown, and cross-account management -- require a separately deployed operational IAM role (the cross-account / home-account role). You deploy it once in this account, and once per additional account you manage, with the cross-account submodule -- it reads every policy document from the running gateway at apply time, so there is nothing to fill in by hand. Without it, most features will not work. See Cross-account / home-account IAM role in the User Guide for the full setup.
Where configuration lives: All gateway configuration is stored in AWS SSM Parameter Store inside your account. There is no external database and no vendor-hosted state.

3. Choose a deployment

Terraform is how Cloud Spectra Gateway is deployed. One module, one apply, and terraform destroy takes it all away again. The standalone AMI below is for a quick look, not for running anything.

MethodBest forHow
TerraformEveryone -- this is the supported pathThe cloudspectra-gateway-aws/aws module plus the cloudspectra/gateway provider
Standalone AMIA 5-minute look, or an air-gapped evaluationLaunch the AMI from the EC2 console; boots with NAT + dashboard

New VPC vs existing/BYO VPC -- one variable

There is one root module and one knob, vpc_id, which decides three ways:

  • Leave it empty (the default) -- Cloud Spectra creates a fresh VPC, subnets, route tables, and an internet gateway for you, allocating the per-AZ subnet CIDRs out of vpc_cidr. The simplest starting point.
  • Set it to vpc-0abc... -- Cloud Spectra deploys into a VPC you already operate, discovers its attached internet gateway, and carves its subnets out of free space. Use this for production accounts where the network is already established.
  • Set it to a VPC name -- anything that is not a vpc- id is matched against the VPC's Name tag and resolved to an id. The same existing-VPC path.
Generate your Terraform: deploy.html is a guided form over every input this module takes, with a live architecture diagram. Fill it in, download main.tf, and you have the file the walkthrough below assumes. Everything you leave alone keeps its module default, so the file stays short.
Standalone AMI path: You can also launch the Cloud Spectra AMI directly from the EC2 console, with no Terraform at all. It boots with sNAT and the dashboard already running, and it is the right way to spend five minutes on the product before deciding. It gives you one instance with no autoscaling, no per-AZ data plane and nothing in version control, so it is not a way to run the gateway. See the standalone walkthrough in the User Guide.

4. Deploy with Terraform

Estimated time: 8-10 minutes

1Subscribe in AWS Marketplace

Visit the AWS Marketplace and subscribe to the Cloud Spectra Gateway listing, then deploy. One Marketplace product per architecture ships the same superset AMI with every feature in it, so there is no tier to buy: the tier you set selects which usage meter your hours post to (Standard or Advanced), and nothing is withheld either way.

2Get the module and your main.tf

The cloudspectra-gateway-aws module ships with your subscription. Put it beside your root configuration, or point source at your own module registry. For the main.tf itself, the quickest route is deploy.html: a guided form over every input, with a live diagram, that hands you a finished file.

3Choose new-VPC or existing/BYO-VPC

Leave vpc_id empty to have Cloud Spectra build the network for you, or set it to a VPC id or Name to deploy into one you already run (see section 3). When in doubt, leave it empty.

4Set the inputs

Every variable has a sensible default, so a working file is short. The ones worth reviewing:

VariableWhat it controls
vpc_idEmpty creates a VPC; a vpc- id or a VPC Name deploys into an existing one.
availability_zonesThe AZs to deploy into, e.g. ["a", "b"]. One ASG is created per AZ; additional AZs add high availability.
instance_typeThe EC2 instance type for gateway nodes. Can be changed live later.
client_cidrThe source CIDR allowed to reach the dashboard and management SSH. Restrict to your office or VPN range.
laneThe Marketplace lane you subscribed to -- developer, enterprise or byol. It picks the listing whose AMI is launched.
gateway_configuration_yamlThe whole-config escape valve: any schema field, including ones with no dedicated variable. deploy.html writes this for you.
Pick the instance for your traffic, not for a feature gate: no feature has a minimum instance size. The inline data-plane services -- Suricata IDS/IPS, antivirus scanning and the AI LLM proxy -- are CPU-work, so their throughput scales with vCPU and a busier link wants a bigger shape. The Enterprise lane sells nothing below 4 vCPU, so on that lane the smallest shape you can pick is already an xlarge. Match the instance architecture to the AMI you subscribed to.

5terraform init && terraform apply

Read the plan, then confirm. Provisioning takes roughly 8-10 minutes. terraform init fetches the cloudspectra/gateway provider from the public Terraform Registry; no mirror configuration is needed.

What the apply builds: a per-AZ Auto Scaling Group (one ASG per Availability Zone, each egressing through its own ENI to avoid cross-AZ data charges), a Gateway Load Balancer (GWLB, GENEVE) for horizontal scale, and an Elastic IP for a stable endpoint. The internals are described in the Architecture guide.

5. First login

6Find the dashboard URL in the outputs

terraform output dashboard_url. It is built from the gateway's Elastic IP, which is your stable dashboard endpoint. Open it over HTTPS on port 9443 -- the dashboard_url output already includes the port. (The dashboard is served by nginx using a per-instance self-signed certificate generated at first boot; an ACM certificate is not used here.) Port 443 is deliberately left free for TLS termination in front of your own services; if you are not using a TLS load balancer, you can set cloudspectra_dashboard_port to 443 to serve the dashboard on the standard HTTPS port.

7Retrieve the initial admin password

Log in as user admin. The one-time initial password lives only in a root-only file on the instance, so you retrieve it over SSH. The stack generated the key pair for you: run the AccessInstancePrivateKeyCommand stack output to download the private key from SSM Parameter Store, then SSH with it. Password authentication and root login are disabled on the appliance, so you authenticate with that key:

# 1. Download the stack-generated private key
#    (copy this command verbatim from the AccessInstancePrivateKeyCommand stack output)
aws ssm get-parameter --name /ec2/keypair/<KEY_PAIR_ID> --with-decryption \
  --query Parameter.Value --output text > ~/.ssh/<KEY_NAME> && chmod 400 ~/.ssh/<KEY_NAME>

# 2. SSH in as user "admin" with that key
ssh -i ~/.ssh/<KEY_NAME> admin@<ELASTIC_IP>

# 3. Read the one-time dashboard init password (root-only file)
sudo cat /etc/cloudspectra/dashboard_init_pw

The initial password is never written to SSM Parameter Store or anywhere else -- this root-only file on the instance is the only copy until you set a permanent password. (Only the SSH private key is stored in SSM, at /ec2/keypair/<KEY_PAIR_ID>.)

8Set a permanent password

The dashboard forces a password change on first login. After you set your own password, the one-time init password stops working. Keep your new password somewhere safe -- the gateway has no vendor-side recovery path because there is no vendor control plane.

Certificate warning: In v1 the dashboard always serves a per-instance self-signed certificate, so your browser will show a warning on first access. This is expected, and the ACM certificate parameter does not change it -- that parameter applies only to TLS termination on port 443 in front of your own services.
Roadmap. Built-in Grafana fleet dashboards -- a single pane of live CPU, memory, disk, and network metrics for every instance across all Auto Scaling Groups -- are on the roadmap and are not enabled in v1. In v1, use the dashboard's own status panels (section 8) to check per-AZ instance health. This is distinct from the ElasticApps Grafana application, which ships today and is one of the stateful apps you can run elastically.

6. NAT works out of the box

As soon as the stack reaches CREATE_COMPLETE, source NAT is already running on the appliance -- no configuration step is required. Private instances whose route tables point default traffic at the Cloud Spectra ENI immediately reach the internet through the gateway's Elastic IP.

graph LR
    subgraph VPC["Your VPC"]
        PRIV["Private subnet instances
(no public IP)"] GW["Cloud Spectra Gateway
sNAT on per-AZ ENI"] end EIP["Elastic IP
(stable egress)"] NET["Internet"] PRIV -->|"0.0.0.0/0 route"| GW GW --> EIP EIP --> NET style GW fill:#d1fae5,stroke:#10b981,color:#065f46 style EIP fill:#dbeafe,stroke:#3b82f6,color:#1e3a8a

9Verify outbound from a private instance

SSH into any EC2 instance in a private subnet that routes through the gateway, and confirm it can reach the internet:

# From a private instance routed through Cloud Spectra:
curl -s https://checkip.amazonaws.com

# The returned IP should match the gateway's Elastic IP
# (the AccessElasticIp value from the stack Outputs).
Multi-AZ tip: Point each AZ's private subnets at that same AZ's Cloud Spectra ENI. Because each AZ egresses through its own ENI, keeping traffic within the AZ avoids cross-AZ data transfer charges. The route-table details are in the User Guide.

7. Enable a feature

Every additional capability is turned on the same way: from the dashboard (or the config API on port 8080, or the Terraform provider). Per-feature configuration reference -- including firewall rules, the Squid proxy, TLS termination, and the AI Gateway -- lives in the User Guide. Here is the pattern using one such feature, the forward HTTP proxy.

flowchart LR
    A["Dashboard (HTTPS 9443)"] --> B["Toggle a feature on"]
    B --> C["Config saved to
SSM Parameter Store"] C --> D["Gateway nodes reconcile
and apply the change"] D --> E["Feature live
on all AZs"] style A fill:#dbeafe,stroke:#3b82f6,color:#1e3a8a style E fill:#d1fae5,stroke:#10b981,color:#065f46

10Enable the forward HTTP proxy (Squid)

  1. Open the dashboard over HTTPS and sign in.
  2. Go to the Web Proxy panel ("Web Proxy Gateway").
  3. Enable the proxy, set the listening port, and optionally configure response caching, domain filtering, and bandwidth limits.
  4. Copy the auto-generated proxy credential from the panel's Proxy Credential card. The username is always admin; the password is generated for you and can be re-issued with Rotate password.
  5. Save. The change is written to SSM Parameter Store and the gateway nodes apply it automatically.

Then point a client at the proxy:

# Send outbound traffic through the Cloud Spectra forward proxy
# (username is always "admin"; copy PROXY_PASSWORD from the
#  Web Proxy panel's Proxy Credential card)
export http_proxy=http://admin:<PROXY_PASSWORD>@<GATEWAY>:<PROXY_PORT>
export https_proxy=$http_proxy
curl -s https://checkip.amazonaws.com
The same pattern applies to every feature. Each of these is enabled from the dashboard with its own settings; see the User Guide for the full per-feature reference.
  • Standard TLS termination (ACM), the IPVS L4 load balancer, destination NAT / port forwarding, the sticky secondary ENI (a pre-provisioned ENI with a fixed private IP that survives instance replacement), and the per-slot route table.
  • The sticky EBS data volume is available in every deployment, like every other feature. See the lineup above.
  • Advanced the Suricata IDS/IPS, nftables firewall rules, domain and URL-category filtering, ET Open threat detection, community threat feeds (abuse.ch, CINS Army, Spamhaus DROP), TLS inspection, antivirus scanning, L7 App-ID, and SIEM export to Splunk, Sentinel, or QRadar.
  • Advanced the AI LLM proxy with its response cache, the semantic cache, and the AI Assistant -- an in-dashboard assistant, backed by Amazon Bedrock, that answers questions about gateway status, configuration, and troubleshooting. It runs in your account; no vendor control plane sees your data.
TLS inspection first for HTTPS scanning. Antivirus scanning inspects downloads that pass through the forward proxy using ClamAV over ICAP. Because HTTPS traffic is encrypted end-to-end, scanning HTTPS downloads requires TLS inspection (SSL bump) to be enabled first -- without it, only cleartext HTTP downloads are scanned. TLS inspection uses a self-managed certificate authority by default, and the keys never leave your account. See the User Guide for the CA trust-distribution steps.
Sizing. Suricata IDS/IPS, antivirus scanning and the AI LLM proxy run inline in the data plane, so their throughput scales with vCPU (see section 4). They start on any shape your lane sells; if inspection is not keeping up with your traffic, resize the gateway.

Key ports

PortService
9443HTTPS management dashboard (nginx, per-instance self-signed certificate)
443TLS termination (HAProxy + ACM) Standard
8080Configuration API
8090AI Gateway, OpenAI-compatible endpoint Advanced
configurableSquid forward proxy port
80HTTP redirect to HTTPS

8. Verify

A quick end-to-end check confirms the gateway is healthy and routing correctly.

CheckHowExpected result
Outbound NATcurl -s https://checkip.amazonaws.com from a private instanceReturns the gateway's Elastic IP
Dashboard reachableOpen the Elastic IP over HTTPS (port 9443)Login page loads
Status panelsSign in and open the dashboard overviewPer-AZ instances healthy; enabled features show active
Forward proxy (if enabled)curl through the proxy portRequest succeeds via the proxy

If outbound traffic does not return the gateway's Elastic IP, confirm the private subnet's route table sends 0.0.0.0/0 to the Cloud Spectra ENI for that AZ. Common questions and troubleshooting are collected in the FAQ.

Before tearing down: If you pointed existing route tables at Cloud Spectra ENIs, revert those routes before terraform destroy to avoid losing connectivity for instances that depend on the gateway.

9. Next steps

You now have a running Cloud Spectra Gateway with NAT live and a feature enabled. Where to go from here:

DocumentWhat it covers
User GuidePer-feature configuration for all of it: sNAT/dNAT, IPVS L4 load balancing, TLS termination (ACM), the sticky secondary ENI and per-slot route table, sticky EBS, per-AZ Auto Scaling, GWLB scaling, the Squid forward proxy, Suricata IDS/IPS and nftables firewall, domain and URL-category filtering, ET Open and community threat feeds, TLS inspection, antivirus scanning, L7 App-ID, SIEM export, the AI Gateway with response caching and the semantic cache, the AI Assistant, and ElasticApps (vertical right-sizing, sticky EBS, and a stable endpoint for a single stateful app). Also the full Terraform and standalone-AMI walkthroughs.
ArchitectureHow the appliance is assembled: the per-AZ ASG model, GWLB/GENEVE data path, Elastic IP endpoint, SSM-backed configuration, the in-account data-plane design, and why there is no vendor control plane.
FAQCommon questions on cost, data residency, security, deployment options, scaling, and troubleshooting.
Recommended reading order: Start with the User Guide to configure the features you need, skim the Architecture guide to understand how it all fits together, and keep the FAQ handy for quick answers.