smolClouds — Your agents build it. SmolClouds runs it. Cloud infrastructure for agent-generated software.
A request in plain language.
Platform log
Apps waking, serving, and going back to sleep.
Every app on the platform spends most of its life stopped. The log is mostly the sound of things falling asleep again.
Illustrative — shows the shape of the event log, not live account data.
connecting…
The lifecycle
An app that costs nothing while nobody uses it.
Between requests the microVM is fully stopped. When a request arrives it is restored from a memory snapshot — already booted, already warm — serves the response, and goes back to sleep.
- 01
Sleeping
No request, no running VM, no compute being billed.
- 02
Restoring
A request arrives and the VM is restored from its memory snapshot.
- 03
Running
The app serves traffic with the state it already had in memory.
- 04
Back to sleep
Idle again — the VM stops and the snapshot stays on disk.
sales-report
No active compute
The problem
Most software does nothing most of the day.
40 apps deployed.
1 running.
Why pay for 40 servers?
- invoice-reportsleeping
- csv-dashboardsleeping
- approval-toolsleeping
- internal-apisleeping
- customer-lookupsleeping
- migration-helpersleeping
- price-calculatorsleeping
- onboarding-checklistsleeping
- slack-digestsleeping
- expense-parsersleeping
- lead-scorersleeping
- pdf-splittersleeping
- inventory-syncsleeping
- qa-checklistsleeping
- release-notessleeping
- webhook-relaysleeping
- shift-plannersleeping
- contract-diffsleeping
- survey-viewersleeping
- uptime-pingersleeping
- budget-trackersleeping
- sku-mappersleeping
- email-parsersleeping
- photo-resizersleeping
- tax-tablesleeping
- ticket-triagesleeping
- meeting-notessleeping
- vendor-listsleeping
- churn-alertsleeping
- log-greppersleeping
- form-buildersleeping
- asset-taggersleeping
- payroll-checksleeping
- seo-auditsleeping
- crm-exportsleeping
- note-searchsleeping
- deploy-botsleeping
- sitemap-gensleeping
- sales-dashboardrunning
- receipt-ocrsleeping
How it works
Three steps, no infrastructure decisions.
01 — Build
- Claude Code
- Codex
- Cursor
- Your code
02 — Deploy
- $ smolclouds deploy
03 — Done
- your-app.smolclouds.com
SmolClouds handles builds, routing, isolation, HTTPS and scale-to-zero.
Built for agents
Built for agents, not dashboards.
Deployment is an API call, not a console session. An agent can ship an app end to end without a human clicking through a control panel.
- YOU
- CLAUDE / CODEX / CURSOR
- SMOLCLOUDSPOST /deploy
- LIVE APP
You: Turn this CSV into a dashboard and share it with the team.
Claude: Building…
Deploying to SmolClouds…
Done: sales-q3.smolclouds.com
Use cases
Perfect for the software nobody wants to host.
Internal Tools
Approval flows, admin tools, team utilities.
Dashboards
Reports, analytics, CSV visualizers.
AI Tools
LLM wrappers, assistants, workflow utilities.
APIs & Webhooks
Low-traffic APIs and event receivers.
Agent Apps
Claude Code, Codex and Cursor generated software.
Temporary Tools
Migrations, one-off utilities, prototypes.
Sleep / wake economics
Zero traffic. Nearly zero compute.
Idle apps are not throttled or scaled down — they are stopped. The cost of an app tracks the time it actually spends serving requests.
No requests
Application
Sleeping
No active compute
Request
Wake
Application ready
Response
Cold wake (p95)
Benchmark pending
Time from request arriving at the edge to the restored app responding.
Warm request
Benchmark pending
Response time while the microVM is already running.
Compute billed while idle
None
A sleeping app has no running VM, so there is no active compute to bill.
Architectural: the microVM is stopped between requests.
We publish a latency figure only once it comes from a benchmark we can reproduce and describe. Until then this space stays empty on purpose.
Architecture
Tiny apps deserve serious isolation.
Small does not mean shared. Each application gets a hardware-virtualised boundary — the same primitive that runs serverless workloads at large cloud providers.
SmolClouds Host
Shared Runtime Snapshot
microVM
App A
microVM
App B
microVM
App C
copy-on-write memory · isolated kernels · private network namespace
Firecracker microVM isolation
Every app runs in its own microVM with its own kernel — not a shared process or container namespace.
Shared warm runtimes
VMs derived from the same base image share memory copy-on-write, so identical runtime pages are stored once.
Sub-second wake architecture
Waking restores a memory snapshot of an already-booted app instead of running install, boot and warm-up.
Developer experience
One command, from repo to URL.
The CLI detects the runtime, builds the app and returns a live HTTPS URL. The same actions are available over the REST API, which is how agents deploy.
- ✓Runtime detection — no Dockerfile required
- ✓HTTPS and a smolclouds.com subdomain by default
- ✓Environment variables encrypted at rest
- ✓Every CLI action available over the REST API
$
Comparison
Where SmolClouds is different.
Other platforms can host these apps — several of them well. The difference is what the platform is optimised for: hundreds of mostly-idle apps deployed by agents.
| Capability | SmolClouds | Traditional cloud |
|---|---|---|
| Agent-first deploy | ✓ | Varies |
| Scale-to-zero | ✓ | Varies |
| App-based UX | ✓ | Usually infra-based |
| microVM isolation | ✓ | Varies |
| Optimized for hundreds of tiny apps | ✓ | General purpose |
| GitHub deploy | ✓ | ✓ |
Pricing
Keep hundreds of small applications deployed.
Pay for the platform and the resources that actually run. Sleeping apps consume no active compute.
Free and Scale plans also available.
View all pricing →Security
Security by isolation.
The isolation boundary is a virtual machine, not a process. That is the same boundary large clouds use for untrusted multi-tenant workloads.
Firecracker microVMs
Each application runs inside its own hardware-virtualised microVM with a separate guest kernel.
Tenant isolation
Applications belonging to different accounts never share a guest kernel or a filesystem.
Encrypted secrets
Environment variables are encrypted at rest and injected into the VM at start, never written into build artifacts.
TLS everywhere
Every smolclouds.com subdomain and every custom domain is served over HTTPS with certificates issued and renewed automatically.
Runtime isolation
Snapshot memory shared between VMs is copy-on-write and read-only from the guest's perspective.
Network isolation
Each microVM gets its own network namespace; inbound traffic reaches the app only through the platform router.
Build as much small software as you want.
Deploy the first one in a couple of minutes. It sleeps when nobody is using it.