Tensor Box Agent Sandbox
Secure and isolated AI execution
A secure, controlled environment for agent code, tools, and file operations
Who is it for
When agents execute code, handle files, or call external tools, the sandbox makes boundaries and security responsibilities explicit.
- Agents run scripts, analyze files, or produce artifacts
- Security teams control network, file, and resource access
- Platform teams replay every execution step
The production loop
Sessions, input files, execution steps, and outputs stay connected for audit and review.
- Isolate runtime and temporary files per task
- Record tool calls, command output, and failure reasons
- Replay abnormal sessions without guesswork
核心能力
为 AI 智能体量身打造的安全沙箱环境
Secure isolation
Container-level isolation gives every task its own runtime and prevents privilege escalation.
Code execution
Safely run Python, Node.js, Shell, and other languages.
Tool calls
Use a rich built-in toolkit and extend it with custom tools.
File system
Secure file read/write and uploads with temporary and persistent storage.
Network controls
Fine-grained allowlists keep network access controlled and data protected.
Resource limits
Limit CPU, memory, disk, and execution time independently.