Policy Enforcement

Policies control which API calls your agents can make. Set per-session when creating a sandbox token.

Default action

policies: {
  default: "deny",  // block everything not explicitly allowed
}

Or:

policies: {
  default: "allow",  // allow everything not explicitly blocked
}

Rules

Rules allow or deny requests matching a domain, method, and path pattern:

policies: {
  default: "deny",
  rules: [
    // Allow all requests to OpenAI
    { domain: "api.openai.com" },

    // Allow only GET to specific GitHub repos
    { domain: "api.github.com", methods: ["GET"], paths: ["/repos/acme/*"] },

    // Allow POST to a specific endpoint
    { domain: "api.stripe.com", methods: ["POST"], paths: ["/v1/charges"] },
  ],
}

Rules are evaluated in order. The first match wins. If no rule matches, the default action applies.

Rate limiting

Limit requests per domain:

policies: {
  default: "allow",
  rateLimit: {
    "api.openai.com": "20/min",
    "api.github.com": "100/hour",
  },
}

Rate limits are enforced per-session. When exceeded, requests are blocked with a 429 response.

Shadow mode

Test policies without enforcing them:

const session = await sylo.createSandboxToken({
  // ...
  shadowMode: true,
});

In shadow mode, policy violations are logged as "would-block" instead of "deny", but the request is allowed through. Useful for testing new policies in production.

Policy webhook

For complex authorization logic, delegate decisions to your own endpoint:

policies: {
  policyWebhook: "https://your-api.com/authorize",
}

Sylo sends the request details to your webhook and blocks or allows based on the response.