Budget Alerts

AI spend can run away quickly. Set up alerts so you don’t wake up to a blown monthly budget because Ada got stuck in a loop overnight.

---

Three-Layer Protection

Set all three:

---

Layer 1: Daily Alert

Purpose: catch anomalies early — don’t wait for a month-end surprise.

Setup

1. Admin Panel → companion tab → Settings → Budget & Alerts
2. Enable Daily Threshold Alert
3. Set the threshold (e.g., $5)
4. Pick a delivery channel:
   • Email (to your registered address)
   • Telegram / LINE (to the bound platform)

[Screenshot: Ada Dashboard → Settings → Budget & Alerts]

Suggested Threshold

Use the 7-day average × 1.5:

7-day average daily spend: $2
→ Alert threshold: $3
→ Any day above that, you get a ping

---

Layer 2: Monthly Hard Cap

Purpose: even if you miss the alert, the budget holds.

Setup

1. Admin Panel → Settings → Monthly Budget
2. Enter the monthly cap (e.g., $50)
3. Pick the action when it’s hit:
   • Soft Stop: AI stops calling external APIs but keeps basic replies
   • Hard Stop: AI fully disables; requires manual reset

Suggested Strategy

Personal:

• Cap at the maximum you’re willing to lose (e.g., $30)
• Use Hard Stop

Business support:

• Cap at 1.5× your budget (for a buffer)
• Use Soft Stop (avoid “no response” for customers)
• Combine with daily alerts to catch anomalies early

---

Layer 3: Provider-Side Limit

Why this layer?

Admin Panel checks are after the fact — spend first, count later. In the worst case (infinite loop), you can burn hundreds of dollars before the check catches up.

Provider-side Hard Limits are preemptive — the API blocks the call. You almost never exceed (occasionally a tiny overage due to billing latency).

OpenAI

1. Sign in to platform.openai.com
2. Top-right avatar → Billing → Usage limits
3. Set:
   • Hard Limit: absolute monthly cap (API refuses above it)
   • Soft Limit: warning email

Anthropic

1. Sign in to console.anthropic.com
2. Settings → Billing
3. Set Monthly Limit

Google AI Studio

Set a Budget Alert in Google Cloud Console:

• console.cloud.google.com → Billing → Budgets & alerts

---

Real-World Setups

Personal (Target: $20/Month)

┌─────────────────────────────────────┐
│ Layer 1 — Daily alert: $1            │
│   Above → Telegram notification      │
├─────────────────────────────────────┤
│ Layer 2 — Admin Panel Hard Stop: $25 │
│   Above → AI auto-disabled          │
├─────────────────────────────────────┤
│ Layer 3 — OpenAI Hard Limit: $30     │
│   Absolute ceiling                  │
└─────────────────────────────────────┘

Small E-commerce Support (Target: $300/Month)

┌─────────────────────────────────────┐
│ Layer 1 — Daily alert: $15           │
│   Above → Email + Telegram          │
├─────────────────────────────────────┤
│ Layer 2 — Admin Panel Soft Stop: $320│
│   Above → AI falls back to simpler  │
├─────────────────────────────────────┤
│ Layer 3 — OpenAI Hard Limit: $400    │
│   Emergency brake                   │
└─────────────────────────────────────┘

---

What Happens When the Cap Triggers

Soft Stop

AI keeps running in a downgraded mode:

• Only the cheapest model
• No proactive tool calls
• Shorter replies

Customers notice the difference, but the line does not go silent.

Hard Stop

AI stops completely. To resume:

1. Admin Panel → Settings → Budget → Reset
2. Or raise the monthly cap
3. Or wait for auto-reset at month start

Urgent need: temporarily raise the provider’s limit directly.

---

Troubleshooting a Spend Spike

If the budget suddenly jumps, open a support ticket or self-diagnose:

1. Usage → 7-day line → find the spike day
2. Switch to Hourly → find the hour
3. Activity → review that hour’s conversations
4. Common causes:
   • AI loop (talking to itself)
   • Abuse (Telegram bot without allowlist)
   • Repeated tool-call retries
   • An expensive model set by mistake

---

Related Docs

• Usage Dashboard Reference
• Cost Optimization
• How Tokens Are Priced