/managed-agents
Managed Agents
Use the official Anthropic SDK and point it at Sandbox0. Keep the Claude Managed Agents resource model while Sandbox0 runs the backend and records traceable session execution.
AI agent infra
Build AI agents
on Sandbox0
Sandbox0 gives AI agents durable workspaces, isolated execution, network control, and credential-safe runtime access. Use Managed Agents for the full backend, or use the Sandbox SDK when you want the lower-level primitives directly.
Prompt for your agent
Read https://sandbox0.ai/llms.txt and help me evaluate Sandbox0 for AI agent infrastructure. Compare Managed Agents vs Sandbox SDK, cite the docs you use, and give me a short integration plan.
Choose your level
Start with the product surface that matches your control needs.
Sandbox0 is one infrastructure stack with two public ways in: a high-level managed agent backend and a lower-level sandbox SDK.
/sandbox
Sandbox SDK
Use sandboxes, templates, volumes, credentials, files, processes, ports, and network policy as direct building blocks.
Runtime guarantees
The infrastructure pieces agents need after the demo.
Agent products tend to grow past stateless model calls. Sandbox0 keeps runtime, storage, network, and credentials under explicit infrastructure control.
state
Persistent workspaces
Keep repositories, caches, artifacts, and checkpoints across sandbox lifecycles.
runtime
Isolated execution
Run commands, tools, app servers, and long-lived sessions inside controlled sandboxes.
policy
Network control
Apply per-sandbox egress and public gateway rules instead of relying on process discipline.
secrets
Credential-safe access
Project supported outbound auth on the egress path so raw secrets do not need to live in agent code.
Build your agent infrastructure on Sandbox0.
Start high with Managed Agents, or drop down to the Sandbox SDK when your architecture needs direct runtime control.