Mis on agendi sandboxing ja milleks see on vajalik?

Agent's sandboxing is an isolated environment where the agent (for example, an LLM agent) operates to restrict its access to system resources and prevent unwanted or dangerous actions.

Why is sandboxing needed:

• Security: prevents execution of malicious code or access to confidential data.
• Resource control: limits the use of memory, CPU, network.
• Error isolation: agent failures do not affect the main system.

For example, if an agent can execute arbitrary code or call external APIs, sandboxing ensures it cannot damage the system or access forbidden data.