LLM agent blue-green deployment cost: budgeting zero-downtime model updates without doubling your AI bill

import runguard
from datetime import datetime, timedelta

# Set a deployment-scoped budget cap
# This cap applies across all agent runs in the green environment
# and resets on each new deployment.
deployment_guard = runguard.DeploymentBudget(
    deployment_id="green-v2.3.1",
    budget_usd=50.00,          # max $50 during validation
    window_hours=4,             # validation window: 4 hours
    on_exceeded="reject_new",   # stop new green sessions, don't kill active ones
)

# In your agent handler:
async def handle_request(user_message: str, session_id: str):
    # Check deployment budget before starting a new green session
    if not deployment_guard.can_start_session():
        # Deployment budget exhausted — route to blue instead
        return route_to_blue(user_message, session_id)

    async with deployment_guard.session(session_id, budget_usd=2.00):
        result = await run_agent(user_message)
    return result

import { DeploymentBudget } from 'runguard';

const deploymentGuard = new DeploymentBudget({
  deploymentId: 'green-v2.3.1',
  budgetUsd: 50.00,
  windowHours: 4,
  onExceeded: 'reject_new',
});

export async function handleRequest(
  userMessage: string,
  sessionId: string
): Promise<string> {
  if (!deploymentGuard.canStartSession()) {
    // Route back to blue environment
    return routeToBlue(userMessage, sessionId);
  }

  return deploymentGuard.withSession(sessionId, { budgetUsd: 2.00 }, async () => {
    return runAgent(userMessage);
  });
}