Local runtimes

Where a sandbox runs changes the business case.

GlobalStacks centralizes orchestration without forcing every agent workload through rented cloud compute.

The default cloud pattern

Easy to start. Expensive to leave on autopilot.

Agent sandboxes are not just web requests. They clone, build, test, wait, retry, stream logs, and sit close to private systems. If every session becomes rented cloud compute, the runtime bill becomes part of every workflow.

Margin

Do not turn every agent session into rented compute when the workload is steady.

Control

Keep runtime execution inside customer-approved hosts, networks, and storage paths.

Distance

Run near repositories, build caches, databases, and internal services.

Placement run

sandbox.create()

policy matched
Signals
Workload
steady agent loop
Code
repo nearby
Cache
blueprint ready
Network
private service fit
Policy engine
lowest_cost
locality wins
Owned host
selected
dev-build-02
Same network
eligible
private systems
Cloud pool
fallback
burst demand
Sandbox request
agent workload
Runs on dev-build-02
owned host with local cache
$ gstacks sandboxes create --placement-policy lowest_cost
effective policy: lowest_cost
signals: byo source, local cache, private-network fit

Before

One sandbox. Four meters.

always billable
remote cloud runtime metering active
Compute
runtime hours
+ cost
Storage
scratch + volumes
+ cost
Network
private traffic
+ cost
Egress
data movement
+ cost
The scheduler is simple. The bill is not.

With GlobalStacks

Place the runtime before the meter starts.

policy first
1
Workload
tests, builds, agents
2
Signals
repo, data, policy
3
Decision
owned, local, cloud
4
Runtime
selected host
owned host
steady work
same network
private systems
cloud pool
burst only
Same control plane. Different runtime economics.

Not anti-cloud

Cloud is a placement option, not the default tax.

Use cloud capacity when elasticity is the point: bursty demand, managed regions, or teams without available hosts. Use local runtimes when execution is predictable, sensitive, close to private systems, or already backed by hardware you operate.

GlobalStacks keeps identity, lifecycle, placement, logs, access, and networking in one control plane. Runtime execution stays movable across connected hosts, Proxmox nodes, local labs, private clusters, and approved managed pools.

The operating model

Start with the machines you trust. Add cloud when the workload asks for it.