Blog Platform
GlobalStacks: infrastructure should start local and move global
Software infrastructure begins with an unnecessary surrender.
A founder has a laptop, perhaps a spare machine or server, but the accepted first step is renting infrastructure from a global provider. The cloud makes deployment easier, so ownership and operations are treated as the same problem.
GlobalStacks begins with a different question:
What if infrastructure could start on a machine you own, acquire the operational properties of cloud and move outward only when the workload requires it?
Why sandboxes are the beginning
Running arbitrary workloads requires isolation before convenience.
A useful platform cannot assume that every application is safe, well behaved or correctly configured. Sandboxes create the boundary within which local machines, remote servers and cloud capacity can participate without turning one workload into control of the host.
Isolation is not an implementation detail.
It is the condition that makes heterogeneous, user-owned infrastructure governable.
The control plane and the workload
The control plane should describe desired state, identity, placement and policy.
The workload should remain portable enough to run wherever those requirements can be met.
Separating the two allows a machine at home, a colocated server and a cloud instance to belong to one operating model without pretending they have identical reliability or trust.
The goal is not hiding differences.
It is making placement a decision rather than a rewrite.
Can locally owned infrastructure meet production requirements?
Home and edge hardware are unreliable.
Networks fail. NAT complicates access. Disks differ. Security and support become harder across uncontrolled environments. Cloud providers solved many of these problems for good reasons.
Correct.
GlobalStacks must not promise that a laptop equals a data centre.
It should expose guarantees honestly, replicate state according to consequence, preserve management sessions through agent upgrades and make promotion to stronger infrastructure routine.
The value is a gradient, not equivalence.
Start local, move global
A workload should be able to mature through stages:
- A sandbox on hardware already available.
- Secure exposure through outbound tunnels.
- Encrypted connectivity between locations.
- Replication and backup.
- Promotion to owned server or colocation.
- Cloud bursting for exceptional demand.
- Global placement when customers justify it.
The application keeps its identity and operational model while its placement changes.
Why build this now
AI has expanded what a single person can build while making compute, data and supplier dependency more important.
Small teams need infrastructure that matches their new capability without forcing enterprise cost at the first experiment. They also need a path to stronger guarantees that does not require abandoning the system that helped them begin.
GlobalStacks is an attempt to separate cloud ergonomics from permanent cloud tenancy.
It will require difficult work: sandbox design, networking, volume attachment, agent recovery, control-plane consistency and honest security boundaries.
Those details are the product.
The series behind the system
The story should be told through the decisions:
- why sandboxing is foundational;
- which isolation guarantees matter;
- why the implementation language matters;
- how volumes remain portable;
- what the control plane owns;
- why terminal sessions survive upgrades;
- why transport choices changed;
- where open source helps or harms the model;
- how a local workload becomes globally reachable;
- when cloud capacity becomes the right destination.
This is not a series about rejecting the cloud.
It is about recovering a choice the cloud made easy to forget.
Infrastructure should be able to begin with what you have, grow into what you need and remain yours while moving between them.
Start local.
Move global.
Do not make the first server the last architectural decision.