StateSet Sandbox
Run AI agents in production with isolated execution, reliable file I/O, and predictable costs.
What it is
Production runtime for agent workflows
StateSet Sandbox is a Kubernetes-backed runtime that gives each agent its own secure sandbox. Use a simple API to execute commands, read/write files, and store artifacts without running your own workers.
Why it matters
Execution you can ship to production
- Shared runners are risky and hard to govern.
- Local setups are brittle and do not scale.
- Agent workflows need reliable execution and clean I/O paths.
How it works
1. Create
Spin up a sandbox in seconds.
2. Execute
Run commands and tools safely.
3. Write
Base64-safe file writes.
4. Read
Fetch files reliably.
5. Store
Upload artifacts for reuse.
6. Stop
End sessions when work is done.
Built for Claude Code
Claude Code CLI ships in the sandbox image with BYOK support.
- Run agent tasks with `claude -p` inside a sandbox.
- Use MCP integration to expose sandbox tools to Claude.
- Bring your own Anthropic API key (BYOK).
Proof and benchmarks
Benchmark harness included
Run the k6 script to measure cold starts and exec latency in your stack.
Request benchmark dataHardened runtime defaults
gVisor isolation, non-root execution, and strict path validation.
Read the FAQReliable file workflows
Base64-safe file I/O and artifact storage for agent outputs.
Get API accessKey features
Isolated pods per agent
Each sandbox runs in its own Kubernetes pod with timeouts and resource limits.
Reliable file I/O
Base64-backed read/write paths avoid WebSocket hangs and binary corruption.
Fast provisioning
Warm pools keep execution responsive for agent loops.
Artifacts and checkpoints
Persist outputs or restore entire workspaces across runs.
API + SDKs
Node and Python SDKs plus a CLI and dashboard for visibility.
Deployment options
Hosted API
Fastest way to get started
Use the StateSet-managed API for quick onboarding and automatic upgrades.
Self-hosted
Bring it to your cluster
Deploy the controller on your Kubernetes cluster to control networking and data.
Pricing summary
Hobby
Free tier for experimentation.
Pro
Higher limits for production workloads.
Team
Volume discounts and advanced controls.
Launch your agent runtime
Go from prototype to production with a secure, predictable execution layer.
Ready to ship agent workflows?
Create an account, generate an API key, and launch your first sandbox in minutes.