sa
Sandbox
Claim isolated runtimes from templates, pause and resume work, expose ports, and run processes.
Sandbox SDK
The sandbox layer
for AI agents
Use the Sandbox SDK when you want direct control over the runtime primitives: sandboxes, templates, volumes, credentials, files, processes, ports, webhooks, and network policy.
Prompt for your agent
Read https://sandbox0.ai/llms.txt and plan a Sandbox SDK integration for my AI agent backend. Cover sandboxes, volumes, files, processes, network policy, credentials, tests, and risks.
SDK workbench
direct API
const sandbox = await client.sandboxes.claim("agent-dev", {
volumes: ["repo-cache"],
network: { egress: "policy" }
});
await sandbox.process.exec("npm", ["test"]);template
Define the runtime
Images, resources, warm processes, defaults, and policy live in reusable templates.
sandbox
Claim execution
Run commands, REPLs, file operations, app servers, and agent tools inside isolated sandboxes.
volume
Persist state
Mount, snapshot, restore, and fork durable storage independently from sandbox compute.
Direct primitives
Build your own agent backend on top of the runtime layer.
The Sandbox SDK is for teams that want to own orchestration, but do not want to rebuild isolated execution, storage, and network control from scratch.
te
Template
Package runtime images, resources, warm pools, volume mounts, and default policy.
vo
Volume
Persist repositories, caches, checkpoints, and artifacts with snapshot, restore, and fork workflows.
cr
Credential
Resolve and project outbound auth without placing raw secret material inside the agent process.
SDKs
Control Sandbox0 from the language your backend uses.
Use Sandbox0 through SDKs and the CLI when you want to assemble your own session model, scheduler, workflow engine, or product-specific agent backend.
go
Go SDK
The preferred hand-written wrapper layer over the generated API client for backend services.
js / py
JavaScript and Python
Use generated SDK surfaces for app servers, automation scripts, and agent-side tooling.
cli
s0 CLI
Use the CLI for local workflows, debugging, and operational access to sandboxes and volumes.
When to choose it
Use the Sandbox SDK when Managed Agents is too opinionated.
Some teams need to own the queue, session schema, approvals, tenant routing, or workflow graph. Sandbox0 gives them the execution substrate directly.
custom control plane
Bring your own orchestration
Build your scheduler, workflow engine, or product-specific session model above Sandbox0 primitives.
deep runtime control
Own process and file behavior
Run the exact command model, file pipeline, context lifecycle, and port exposure your agent needs.
infra boundaries
Keep policy explicit
Control templates, volume lifecycle, credential sources, egress rules, and cluster deployment shape.
Use Sandbox0 as your agent runtime substrate.
Start with sandbox docs when you want direct runtime control, or use Managed Agents when you want the higher-level session API.