Hardened Sandbox
The Hardened Sandbox is OmniIDE's security perimeter. Every command executed by the Omni-Agent runs inside this isolated environment, protecting your host operating system from unintended side effects.
Why a Sandbox?
Autonomous agents execute code without human review at every step. This introduces risk — a malformed command could delete files, install malware, or corrupt system state. The Hardened Sandbox ensures that agent operations are contained.
Security Boundaries
| Boundary | Policy |
|---|---|
| File System | Agent can only read/write within the current workspace |
| Network | Configurable — disabled by default for sandbox commands |
| Process Execution | Commands run in an isolated process tree |
| Environment | Sandboxed environment variables — no host env leakage |
| Timeout | Commands are killed after configurable timeout (default: 30s) |
Configuration
{
"omni.sandbox.enabled": true,
"omni.sandbox.timeout": 30000,
"omni.sandbox.allowNetwork": false,
"omni.sandbox.allowFileWrite": true,
"omni.sandbox.allowedPaths": ["./src", "./tests"]
}Warning: Disabling the sandbox (
omni.sandbox.enabled: false) allows the agent to execute commands with full system access. Only do this if you fully trust the agent's output.Related
- Omni-Agent — The autonomous engine
- Security Model — Full security architecture
- Sandbox Configuration Guide — Advanced sandbox setup