Open source infrastructure for people and agents

You built it.
Now run it.

Put your first website on an independent cloud, understand how it works, then grow from one useful node to a resilient Docker Swarm platform.

human + agent / one CLI

Your briefPlan

$ pnpm run doctor -- --json

{ "ok": true, "passed": 18, "failed": 0 }
PLAN READYThe workstation and local inputs are ready.

Illustrative handoff · commands stay visible · actions follow your authorization

OPENSTACKHEATANSIBLEDOCKER SWARMTRAEFIK

Two ways in · one operating model

Learn every layer.
Or delegate with confidence.

Manual mode turns the deployment into a course. Agent mode compresses the same work into a reviewed plan and verified execution. Nothing important becomes a black box.

Checkpoint
Manual learningYou type. TOAD explains.
Agent-assistedYou decide. The agent operates.
Understand

Read the architecture and identify what Heat, Ansible, Swarm, and Traefik each own.

The agent reads the operating contract and explains the same ownership boundaries before acting.

Prepare

Create the cloud user, place clouds.yaml, and run every local prerequisite check yourself.

Provide the protected credential file; the agent diagnoses readiness without asking you to paste secrets.

Plan

Inspect the Heat environment, chosen image, network, service shape, and expected cost line by line.

Receive a compact plan with the exact stack, resources, changes, cost, and stop points.

Deploy

Type the declarative command and follow Heat, host configuration, Swarm, ingress, and monitoring.

Approve the displayed plan once; the agent reconciles the same declared files and narrates progress.

Prove

Read the verifier output and account for each node, replica, certificate, endpoint, and boundary.

Get the same machine-readable evidence summarized with failures, limits, and the smallest next action.

Same declared state · same safety checks · same acceptance evidence

What it gives you

Cloud primitives,
shaped into a platform.

TOAD keeps the infrastructure understandable while adding the contracts a maintained service needs.

01

A cloud you can leave

OpenStack resources are declared in Heat, configured by Ansible and operated through a TypeScript CLI. There is no proprietary control plane to keep alive.

  • Heat
  • Ansible
  • TypeScript
02

Routing that feels native

Traefik discovers Swarm services from labels, obtains HTTPS certificates and sends traffic through a shared overlay—without publishing application ports.

  • Traefik
  • Automatic HTTPS
  • Swarm labels
03

Operations with evidence

Machine-readable diagnosis, external probes, metrics, logs, alerts and declared acceptance checks make failures explainable to people and agents.

  • Prometheus
  • Grafana
  • Loki
04

Guardrails by default

Immutable images, resource limits, rollback policy, encrypted backups and mTLS-protected administration are enforced before a service is deployed.

  • mTLS
  • age
  • Policy checks

One contract, two scales

Start with what you need.
Keep the path forward.

A profile changes the topology, not the operating model. The manifests, security policy and application workflow stay familiar as the cloud grows.

Small but complete

One useful node

The same routing, certificates, monitoring and application contract on the smallest practical footprint.

  • 1 manager
  • Direct ingress
  • Best for evaluation and small services
Internet
Public IP
Manager 1
Traefik
Apps
Observe

A closed operational loop

Declare. Reconcile.
Prove. Repeat.

Every normal change returns to declared state and ends with evidence. Agents can follow the same narrow runbooks as a human operator.

  1. 01doctor

    Prove the workstation

    Check tools, credentials and safe local state without exposing secret contents.

  2. 02apply

    Reconcile the platform

    Create or update cloud resources, hosts, Swarm, ingress, DNS and monitoring.

  3. 03app deploy

    Ship declared services

    Validate policy, deploy with rollback and run the application’s HTTPS checks.

  4. 04verify

    Keep the receipt

    Measure quorum, replicas, DNS, TLS, routes and monitoring in one JSON report.

Before an application reaches the cluster

Immutable imageNo public portsMemory boundedRollback declaredHTTPS checkedBackups explicit

Built for agents · clear for people

Share the evidence.
Keep the authority.

People and coding agents inspect the same contexts, replicas, task errors, endpoint checks, and operation history. Sensitive actions still stop for an explicit human decision.

A useful agent brief

“Diagnose my-service. Show me the failed check and the smallest declarative fix. Do not deploy until I approve.”
See the agent operating contract →
{
  "app": "my-service",
  "ok": false,
  "replicas": "1/2",
  "https": 503,
  "next": "inspect failed task"
}

Show the receipts—and the limits

Operational claims
need evidence.

TOAD distinguishes a recorded result from an architectural promise. Recovery procedures state exactly what was restored and what remains to be tested.

RECORDED

Declarative rebuild

Cloud resources, Swarm, ingress, monitoring, and routes can be reconciled from the repository and protected operator inputs.

RECORDED

Encrypted recovery

Platform volumes have bounded age-encrypted backup and exact-volume restore procedures.

BOUNDARY

Local state is not replicated

Three managers protect quorum, not application data. Stateful services need an explicit storage and recovery profile.

Choose deliberately

Swarm is a feature
when simplicity matters.

TOAD is not trying to imitate every platform. It chooses a compact operating surface and excellent Traefik integration for teams that want to own the result.

ConsiderationTOAD / SwarmKubernetesManaged PaaS
Starting footprint1 managerControl plane + workersProvider-defined
Operational surfaceSmallLargeSmall
Infrastructure ownershipYour OpenStack projectYour clusterProvider account
Application routingTraefik labelsIngress resourcesPlatform-specific
Exit pathHeat + Compose + backupsCluster manifests + backupsVaries by provider

Kubernetes remains a possible separate provider—not a hidden mode that makes Swarm harder to understand.

GNU AGPL-3.0-or-later

Your infrastructure should
remain yours.

Use it, study it, adapt it for clients, and keep the complete system reproducible.