RuntimeeGuides
Best Practices
Best practices for using Runtimee
Best Practices
Follow these best practices when using Runtimee in your autonomous systems.
1. Use Environment Variables for Secrets
Never hardcode your API key or other secrets in your source code.
// Bad
const rt = new Runtimee({ apiKey: "sk_live_1234567890abcdef" })
// Good
const rt = new Runtimee({
apiKey: process.env.RUNTIMEE_API_KEY
})2. Handle All Authorization Decisions
Always handle all three possible authorization decisions: approved, denied, and pending-review.
const { decision, policyResults } = await rt.actors.previewPay(actor.id, { /* ... */ })
switch (decision) {
case "approved":
// Proceed with payment
break
case "denied":
// Handle denial (e.g., notify user, log, etc.)
break
case "pending-review":
// Handle pending review (e.g., flag for manual review)
break
}3. Use Idempotency Keys for Payments
To avoid duplicate payments, use idempotency keys when making payments.
const execution = await rt.actors.pay(actor.id, {
target: "openai:gpt-4-turbo",
amount: "5",
purpose: { type: "llm-inference", id: "chat-001" },
idempotencyKey: "unique-key-for-this-request" // e.g., UUID or request ID
})4. Set Reasonable Budget Limits
Start with conservative budget limits and increase as needed.
// Start with a small budget for testing
const actor = await rt.actors.create({
name: "test-agent",
budget: { amount: "10", currency: "USDC", period: "monthly" }, // 10 USDC
policies: [ /* ... */ ]
})5. Monitor Actor Status Regularly
Regularly check actor status to detect issues early.
// Check status every hour or before critical operations
const status = await rt.actors.status(actor.id)
if (status.status === "frozen") {
// Handle frozen actor
}
if (parseFloat(status.budget.remaining) < 10) {
// Warn about low budget
}How is this guide?