The scariest moment in AI engineering isn't a production outage — it's opening your provider dashboard Monday morning and seeing a four-figure bill from a weekend retry loop.
Budget alerts are your safety net. AgentBurn lets you set per-agent, per-period spending thresholds that fire before you hit your limit, not after.
Why Per-Agent Alerts Matter
A global spending cap is a blunt instrument. If your total monthly budget is $5,000 and your research agent burns through $4,000 in the first week, your support bot goes dark for three weeks. Per-agent alerts let you allocate budget where it matters and catch anomalies at the source.
Alert Configuration
AgentBurn supports three period types: daily, weekly, and monthly. Best practice is to set all three:
- Daily alerts — Catch runaway loops and retry storms within hours
- Weekly alerts — Spot gradual cost creep from prompt changes or traffic spikes
- Monthly alerts — Stay within your overall budget allocation per agent
Webhook Integration
AgentBurn fires webhook notifications when spend crosses your threshold. Point it at Slack, Discord, PagerDuty, or any HTTP endpoint:
# In your .env
ALERT_WEBHOOK_URL="https://hooks.slack.com/services/T00/B00/xxx"The webhook payload includes the agent name, current spend, budget limit, period type, and a direct link to the cost dashboard for that agent.
Setting Effective Thresholds
Start by running your agents for a week with no alerts, just tracking. Look at the daily cost distribution in the AgentBurn dashboard. Set your daily alert at 1.5x the average daily cost — tight enough to catch anomalies, loose enough to avoid noise.
For monthly budgets, take your target annual AI spend, divide by 12, then allocate across agents by priority. Your customer-facing agents should get a larger share than internal tools.