Multi Architecture Provisioning Tool
Spin up cloud machines in seconds. Tear them down just as fast.
Spot pricing. Airgap topologies. CI/CD native. Built for operators.
What is mapt?
mapt is a command-line tool for provisioning and destroying cloud environments across AWS, Azure, GCP, and IBM Cloud. It wraps multi-cloud infrastructure into a single, consistent interface — optimized for cost, speed, and CI/CD integration.
mapt <provider> <target> <create|destroy> [flags]
One pattern. Every cloud. Every OS.
Quickstart
Pull the container and provision a Fedora machine on AWS spot:
podman run -d --name mapt-fedora \
-v ${PWD}:/workspace:z \
-e AWS_ACCESS_KEY_ID=<key> \
-e AWS_SECRET_ACCESS_KEY=<secret> \
-e AWS_DEFAULT_REGION=us-east-1 \
quay.io/redhat-developer/mapt:latest aws fedora create \
--project-name my-fedora \
--backed-url file:///workspace \
--conn-details-output /workspace \
--spot
Connection details land at ${PWD}/host, ${PWD}/username, and ${PWD}/id_rsa. Destroy with the same flags, swapping create for destroy.
What can you provision?
Instances
Services
| Service |
AWS |
Azure |
GCP |
IBM Cloud |
Description |
| Kind |
docs |
docs |
— |
docs |
Lightweight Kubernetes via Kind |
| EKS |
docs |
— |
— |
— |
Managed Kubernetes with spot node groups |
| AKS |
— |
docs |
— |
— |
Managed Kubernetes |
| OpenShift SNC |
docs |
— |
— |
docs |
Single-node OpenShift for testing |
| Mac-Pool |
docs |
— |
— |
— |
Shared Mac host pool — amortize the 24h minimum |
Architectures
| Architecture |
Providers |
| x86_64 |
AWS, Azure, GCP, IBM Cloud |
| arm64 |
AWS, Azure |
| s390x |
IBM Cloud |
| ppc64le |
IBM Cloud |
Key features
Spot-optimized provisioning
mapt scans placement scores and pricing across all regions to find the best cost vs. availability balance — no manual region hunting. If a region doesn't have the instance you need, mapt falls back automatically. On GCP, spot selection uses the beta Capacity Advice API for obtainability, preemption rate, and pricing scoring.
mapt aws rhel create --spot \
--project-name my-rhel --backed-url file:///workspace \
--conn-details-output /workspace
Hardware-spec instance selection
Describe the machine you need; mapt picks the right instance type:
mapt azure fedora create \
--arch arm64 --cpus 4 --memory 16 \
--project-name fedora-arm --backed-url file:///workspace \
--conn-details-output /workspace
Flags: --arch, --cpus, --memory, --nested-virt, --compute-sizes
Details: instance selection docs
Airgap topology
Provision an isolated machine behind a jump bastion. mapt wires up the full network — you get bastion connection details alongside the target host.
mapt aws rhel create --airgap \
--project-name rhel-airgap --backed-url file:///workspace \
--conn-details-output /workspace
Outputs: host, username, id_rsa, bastion_host, bastion_username, bastion_id_rsa
Self-destruct timer (serverless mode)
Set --timeout and mapt will tear itself down automatically if the destroy never runs — pipeline crash, lost state, whatever. No orphaned resources, no surprise bills.
Details: serverless mode docs
CI/CD integrations
mapt machines register themselves with your CI system at provision time — nothing to configure after the fact.
GitHub Actions self-hosted runner
mapt aws fedora create --spot \
--install-ghactions-runner \
--ghactions-runner-repo "https://github.com/your-org/your-repo" \
--ghactions-app-id "123456" \
--ghactions-app-installation-id "789012" \
--ghactions-app-private-key "/path/to/private-key.pem" \
--project-name fedora-runner --backed-url file:///workspace \
--conn-details-output /workspace
Auth methods: GitHub App (recommended), PAT, or pre-generated registration token.
Supported targets: AWS (Windows, RHEL, Fedora, macOS) · Azure (Windows, RHEL) · IBM Cloud (Power, Z)
Details: self-hosted runner docs
GitLab Runner
mapt aws fedora create --spot \
--glrunner-token <token> \
--project-name fedora-gitlab --backed-url file:///workspace \
--conn-details-output /workspace
Supported targets: AWS (Windows, RHEL, Fedora, macOS) · Azure (Windows, RHEL) · IBM Cloud (Power, Z)
Details: GitLab runner docs
Tekton tasks
Tekton tasks for dynamic provisioning inside pipelines are available in the tkn/ directory.
Running mapt
Container (recommended)
podman run -d --name mapt \
-v ${PWD}:/workspace:z \
-e AWS_ACCESS_KEY_ID=<key> \
-e AWS_SECRET_ACCESS_KEY=<secret> \
-e AWS_DEFAULT_REGION=us-east-1 \
quay.io/redhat-developer/mapt:latest aws fedora create \
--project-name my-env \
--backed-url file:///workspace \
--conn-details-output /workspace
The --backed-url volume mount holds your stack state — keep it, you need it to destroy.
Binary
go install github.com/redhat-developer/mapt/cmd/mapt@latest
mapt --help
Telemetry
mapt collects anonymous usage statistics to understand adoption: the command path, mapt version, operating system, architecture, and a random installation ID. The data is used only for aggregate statistics and nothing else. mapt does not send command arguments, logs, personal identifiers, or cloud details. A one-time notice is printed on the first run.
Telemetry is enabled by default. To opt out, pass --no-telemetry or set MAPT_TELEMETRY=0 or DO_NOT_TRACK=1 — set the variable in your shell profile to make it permanent, or set "enabled": false in ~/.mapt/telemetry.json to opt out via config file. Deleting that file resets your installation ID.
State management
mapt uses Pulumi under the hood. Stack state is stored at --backed-url:
- Local:
file:///absolute/path — simplest, works for local dev
- S3:
s3://your-bucket — required for serverless mode and shared CI
- Azure Blob:
azblob://your-container
The --project-name flag namespaces stacks, so you can run multiple environments from the same backend.