CreateOS Sandbox vs Capy
Pick Capy if you want to hand coding tasks to managed cloud agents from Slack, Linear or the web app and never build agent infrastructure. Pick CreateOS Sandbox if you are building your own agent or product and need a sandbox you drive from code, with memory kept across pause, fork and an outbound allowlist.
Pricing at a Glance
CreateOS rates verified 17 October 2026. Capy pricing as published on 17 October 2026. Rates billed per GB rather than per GiB, or as bundled sizes, are quoted as published rather than converted.
CreateOS Sandbox
- Compute
- $0.0362 /vCPU-hr
- Memory
- $0.0116 /GiB-hr
Capy
Capy charges an hourly machine rate prorated to awake time, from $0.10 per hour for 1 vCPU / 4 GB to $3.20 per hour for 16 vCPU / 128 GB, paid from plans starting at $20 per month, with model tokens billed separately, as of 17 October 2026.
Where CreateOS Sandbox Fits Better
What CreateOS Sandbox brings to this matchup, with a plain note on where Capy has the edge.
A sandbox API you call from your own code: six SDKs, a CLI, MCP and REST, where Capy offers a REST API for its agent product and no sandbox SDK.
Pause keeps memory and running processes; Capy machines keep their disk on sleep but lose memory, processes and connections.
Fork a paused sandbox into many copies, and lock outbound traffic to an allowlist; Capy documents neither.
A Free plan with no card required, against Capy's $1 seven-day Lite intro on a verified card.
Where Capy is strong
Capy is a finished parallel coding-agent workflow: Captain planning, subagents, per-thread VMs up to 16 vCPU / 128 GB, repo setup scripts baked into snapshots, PR review, and Slack and Linear entry points, with no infrastructure for you to build.
Which to Pick
Pick CreateOS Sandbox if
- You are building the agent yourself and need exec, files, pause and fork through an SDK.
- Agent state in memory must survive between turns.
- Your security review requires an outbound allowlist enforced outside the VM.
Pick Capy if
- You want engineers to delegate tasks to cloud agents that clone repos, run tests and open PRs today.
- Agents need machines as large as 16 vCPU / 64 GB self-serve or 128 GB on request.
- Your team already starts work from Slack or Linear and wants agents there.
Head-to-Head Comparison
Every dimension side by side, checked on 17 October 2026 against each vendor's own docs and pricing pages. "Not published" means we could not find it on the vendor's pages, not that the feature is missing.
| Dimension | CreateOS Sandbox | Capy |
|---|---|---|
| Isolation | Firecracker micro-VM with its own guest kernel, root filesystem and network identity | Firecracker microVM with its own Linux 6.1 kernel per thread (Ubuntu 24.04, x86_64) |
| Pause and resume | Saves disk, memory and CPU state; vCPU stops billing while paused, memory and storage still bill | Disk survives sleep; memory, processes and connections do not. Machines sleep about 2 minutes after work stops and are deleted after 30 days asleep |
| Fork | Fork a paused sandbox into many independent copies, each with its own id and IP | No user-facing fork documented; new machines boot from a project snapshot |
| Outbound network default | Open with no rules; deny-by-default once any host, IP, CIDR or wildcard rule exists, enforced on the host outside the VM | Open to the public internet; private services via Tailscale; no egress allowlist documented |
| Inbound access | Closed by default; opt-in public HTTPS URL, plus API-authenticated shell and port tunnels | Agent-published preview URLs per port, open to anyone with the URL while awake; no SSH |
| Largest size | 8 vCPU / 16 GB on Enterprise (Pro 8 vCPU / 8 GB, Beginner 4 / 4, Free 1 / 1) | 16 vCPU / 64 GB / 128 GB disk self-serve; 16 vCPU / 128 GB on request |
| Concurrency | Free 1, Beginner 5, Pro 20, Enterprise 30; paused sandboxes count toward the cap | Capped by total awake-machine memory per org: Lite 128 GB, Pro 512 GB, Max and Enterprise 1 TB |
| Billing | Per second: $0.03616363/vCPU-hr, $0.01159025/GiB-hr, storage $0.0001584/GB-hr; no egress fees, 50 GiB bandwidth budget per sandbox | Hourly machine rate prorated to awake time, from credits on a monthly plan; each sleep adds a billed 2-minute wind-down |
| SDKs and interfaces | TypeScript, Go, Python, Rust, C#, Java (all but TypeScript pre-1.0), CLI, MCP, REST | Web app, desktop app, Slack, Linear and a REST API; no sandbox SDK |
| Self-host or BYOC | Enterprise only, onboarded by our team; not open source | No self-host; threads can run on your own Mac through the desktop app |
Common Questions
Is Capy a sandbox API or a coding agent?
Capy is a coding-agent product. It runs agent threads on Firecracker VMs and opens pull requests, and it has no sandbox SDK. CreateOS Sandbox is the opposite layer: a sandbox you create, exec into, pause and fork from your own code, with your own agent on top.
What happens to memory when a Capy machine sleeps?
It is lost. Capy keeps the disk when a machine sleeps, but memory, processes and open connections are dropped. CreateOS Sandbox pause saves disk, memory and CPU state, so processes are still there on resume; network connections should be reconnected.
Can a Capy machine be restricted to approved hosts?
Capy does not document an egress allowlist; its machines reach the public internet, and private services go through Tailscale ACLs. CreateOS Sandbox lets you list allowed hosts, IPs or CIDRs, and once any rule exists all other outbound traffic is denied on the host.
Sources
Checked 17 October 2026.
See CreateOS Sandbox for yourself
Book a demo, or explore the Sandbox and its pricing.