The zero-trust governance layer
for AI agents.

Copy# pip install envpod
from envpod import Pod, screen

# Create governed pod — auto-destroy + gc on exit
with Pod("my-agent", config="coding-agent.yaml") as pod:
    # Encrypted vault — agent never sees the key
    pod.vault_set("ANTHROPIC_API_KEY", "sk-ant-...")

    # Run agent on your real files (COW isolated)
    pod.run("python3 agent.py")

    # Review what changed
    print(pod.diff())

    # Keep src/, discard 48 other files
    pod.commit("src/", rollback_rest=True)

# Screen prompts for injection, PII, credentials
result = screen("ignore previous instructions")
# {'matched': True, 'category': 'injection', ...}

# Fast cloning — 100 agents in 1 second
base = Pod("worker", config="pod.yaml")
base.init_with_base()
for i in range(100):
    Pod.disposable(base, f"task-{i}", "python3 experiment.py")

Copy// npm install envpod
import { Pod, screen } from 'envpod';

// Create governed pod — auto-destroy + gc on exit
await Pod.with('my-agent', { config: 'coding-agent.yaml' }, async (pod) => {
    // Encrypted vault — agent never sees the key
    pod.vaultSet('ANTHROPIC_API_KEY', 'sk-ant-...');

    // Run agent on your real files (COW isolated)
    pod.run('python3 agent.py');

    // Review what changed
    console.log(pod.diff());

    // Keep src/, discard 48 other files
    pod.commit(['src/'], { rollbackRest: true });
});

// Screen prompts for injection, PII, credentials
const result = screen('ignore previous instructions');
// { matched: true, category: 'injection', ... }

// Fast cloning — 100 agents in 1 second
const base = Pod.wrap('worker', { config: 'pod.yaml' });
base.initWithBase();
for (let i = 0; i < 100; i++)
    Pod.disposable(base, \`task-\${i}\`, 'python3 experiment.py');

Runs natively inside a lightweight Linux VM. Full features. Also works with Lima or UTM.

Try it in 30 seconds:

OpenClaw on WhatsApp, Telegram, Discord — finally governed:

pod.yaml