Simple pricing
for agent sandboxes
Start with usage-based pricing. Scale as your agents grow.
Runra Sandbox is priced around the resources your agents actually use: sandbox runtime, memory, storage, pause snapshots, and network egress.
Developer
For individuals and small projects building agent prototypes.
Best for
- Personal projects
- Agent prototypes
- Internal demos
- Early-stage experiments
Startup
For teams building production agent features.
Best for
- AI coding products
- Agent SaaS products
- Internal automation
- Multi-user agent workspaces
- Production pilots
Growth
For teams running agents at larger scale.
Best for
- Agent platforms
- Developer tools
- Multi-tenant SaaS
- High-concurrency workloads
- Long-running agent sessions
Enterprise
For organizations that need custom scale, compliance, deployment, or support.
Best for
- Enterprise agent platforms
- Regulated environments
- High-scale workloads
- Private infrastructure requirements
- Teams that need dedicated sandbox capacity
Usage-based overages
Included usage is designed to make pricing predictable. If you exceed your plan limits, additional usage is billed based on actual consumption.
Exact pricing may vary by region and workload profile.
What counts as a sandbox hour?
A sandbox hour measures active sandbox runtime. If a sandbox runs for 30 minutes, it consumes 0.5 sandbox hours.
Paused sandboxes do not consume active runtime, but their workspace and snapshot storage may continue to count toward storage usage.
Active sandbox
An active sandbox is running and can execute commands, run processes, expose ports, and serve agent workloads.
Consumes runtime
Paused sandbox
A paused sandbox is not executing, but its workspace and runtime state can be preserved for later resume.
No active runtime · May consume storage
Why not price per request?
AI agent workloads are not simple API requests. One agent run may last 10 seconds. Another may last 2 hours, install dependencies, run tests, start a server, pause overnight, and resume later.
Runtime-based pricing maps more closely to the real cost of sandbox infrastructure.
Free trial
Start with free trial credits to test Runra Sandbox with your agent workload. No credit card required for the developer preview.
Compare plans
| Feature | Developer | Startup | Growth | Enterprise |
|---|---|---|---|---|
| Monthly price | $19/ month | $99/ month | $499/ month | Custom |
| Included sandbox hours | 100 | 750 | 5,000 | Custom |
| Max vCPU per sandbox | 2 vCPU | 4 vCPU | 8 vCPU | Custom |
| Max memory per sandbox | 4 GB | 8 GB | 16 GB | Custom |
| Workspace storage | 10 GB | 100 GB | 1 TB | Custom |
| Pause/resume | Basic | Standard | Advanced | Custom |
| Port exposure | Limited | Included | Included | Included |
| Custom images | No | Limited | Included | Included |
| Network policy | Basic | Standard | Advanced | Custom |
| Observability export | Included | Included | Included | Included |
| Support | Community | Priority | Dedicated | |
| Dedicated capacity | No | No | Optional | Included |
| SSO/SAML | No | No | No | Optional |
Frequently asked questions
Is Runra Sandbox the same as Runra?
No. Runra is the open-source runtime framework for running server-side AI agents. Runra Sandbox is the hosted sandbox infrastructure for safely executing those agents in isolated cloud environments.
Is Runra Sandbox open source?
The Runra runtime, SDK, adapters, and observability integrations are open source. Runra Sandbox is the hosted managed service. The underlying sandbox engine is powered by CubeSandbox.
Can I use my own sandbox provider?
Yes. Runra is designed around pluggable sandbox providers. You can use Runra Sandbox, CubeSandbox self-hosted, Docker, E2B, or a custom provider.
Can I use my own observability stack?
Yes. Runra supports exporting structured events to systems like Axiom, OpenTelemetry, JSONL, Postgres, and custom sinks.
Do paused sandboxes cost money?
Paused sandboxes do not consume active runtime, but they may consume workspace and snapshot storage.
Can agents expose ports?
Yes. Runra Sandbox supports controlled port exposure for previews, dev servers, browser-based workflows, and web apps created by agents.
Can I run untrusted code?
Runra Sandbox is designed for isolated agent execution, including code generated or modified by AI agents. However, you should still apply proper resource limits, network policies, and tenant isolation for production workloads.
Can I self-host?
You can self-host the lower-level sandbox engine through CubeSandbox. Runra Sandbox is the managed hosted option for teams that do not want to operate sandbox infrastructure themselves.
What regions are available?
Initial regions will be limited during early access. Enterprise customers can request dedicated capacity or region-specific deployments.
Do you support custom images?
Custom images are planned for paid plans and enterprise deployments. They are useful for teams that need preinstalled dependencies, language runtimes, internal tools, or optimized agent environments.
Can I use Runra Sandbox without the Runra runtime?
Yes. You can use Runra Sandbox directly through the API. The Runra runtime provides a higher-level developer experience, adapter system, observability model, and agent lifecycle controls.
Start running agents in secure sandboxes
Use Runra Sandbox to give every AI agent an isolated, pause-resumable cloud workspace.