If you're running AI workloads on SAP AI Core, you've probably asked yourself at some point: “How much have we actually spent this month so far?” The SAP BTP Cockpit shows you the numbers — but only if you remember to check. And by the time you notice something is off, the month is already over.

This post walks through an open-source tool I built to solve exactly that: AI Core Consumption Monitor — a lightweight Python agent that runs on SAP BTP Cloud Foundry, checks your AI Core capacity unit usage daily, and sends you a notification before you hit your budget.

🔗 Open source on GitHub: github.com/SAP-samples/btp-ai-core-usage-monitor
🔗 AI4U project page: ai4u-website.cfapps.eu10-004.hana.ondemand.com/project/btp-ai-core-usage-monitor

The Problem

Consider Dominik, a BTP Subaccount Administrator. His developers are running RAG pipelines, orchestration workflows, and chatbots — all consuming AI Core capacity units. He has no real-time visibility into how much has been spent. By the time the monthly report arrives, tens of thousands of euros may already be gone. The result: customer shock, escalations, and eroded trust in the platform.

SAP AI Core charges by capacity units (CU) — a normalized metric that aggregates token consumption across all models (Claude, GPT, embedding models, etc.). The usage data is available via the SAP BTP Usage Data Management API, but there's no built-in alerting when you approach a spending threshold.

What the Monitor Does

The agent runs as a background process on BTP Cloud Foundry and:

1. Fetches month-to-date AI Core capacity unit usage from the UAS Reporting API (OAuth2, no manual token management)
2. Aggregates consumption across all models and deployments
3. Compares against your configured monthly spending limit
4. Sends a notification based on the alert level:
   • ✅ Daily update — below warning threshold (configurable via RECEIVE_DAILY_UPDATE_EMAILS, default: on)
   • ⚠️ Warning — at 60% of limit (configurable via WARNING_THRESHOLD_PERCENT)
   • 🚨 Alert — at 80% of limit (configurable via ALERT_THRESHOLD_PERCENT)

The check runs automatically once a day at a configurable time (default: 07:00 UTC, set via CHECK_TIME_UTC). If you only want to be notified when thresholds are crossed — not every day — simply set RECEIVE_DAILY_UPDATE_EMAILS=false. You can also trigger a check on demand via a REST API call — useful for ad-hoc checks or integration into CI/CD pipelines.

What You Get in the Email

• Colour-coded alert banner (green / amber / red)
• Three key metrics: CU used, monthly limit, projected month-end
• Visual progress bar (Outlook-compatible)
• Per-model consumption breakdown table
• Configured threshold reference

The projection is a simple linear extrapolation: if you've used X CU in the first N days of the month, you're on track to use X / N * days_in_month by month end. Not sophisticated, but surprisingly useful for catching runaway consumption early.

Two Notification Channels

Option A: SMTP Email

Works with any SMTP server — Office 365, Gmail, SendGrid, etc. The HTML email is built directly by the app, so you get the full rich layout regardless of your email client.

Option B: SAP Alert Notification Service (ANS)

When ENABLE_ANS=true, the agent posts a structured event to the ANS Producer API. ANS then routes it to whatever channel you've configured — email, Slack, Microsoft Teams, PagerDuty, or any other action in your ANS instance.

The event carries structured tags (subaccountId, percentageUsed, projectedCu, periodStart, periodEnd, colour tokens) that you can use in ANS Payload Templates to build a custom HTML email directly in ANS — without any SMTP server.

Architecture

┌─────────────────────────────────────────────────────────────┐
│  SAP BTP Cloud Foundry                                      │
│                                                             │
│  ┌──────────────────────────────────────────────────────┐   │
│  │  ai-core-monitor (Python / Flask)                    │   │
│  │                                                      │   │
│  │  APScheduler ──► run_monitoring_job() [daily 07:00]  │   │
│  │       │                                              │   │
│  │       ▼                                              │   │
│  │  UAS Reporting API ──► fetch usage (OAuth2)          │   │
│  │       │                                              │   │
│  │       ▼                                              │   │
│  │  Aggregate ai-core capacity_units                    │   │
│  │       │                                              │   │
│  │       ▼                                              │   │
│  │  Compare vs. SPENDING_LIMIT                          │   │
│  │       │                                              │   │
│  │       ├── < 60%  → INFO  (daily email if enabled)    │   │
│  │       ├── ≥ 60%  → WARNING email                     │   │
│  │       └── ≥ 80%  → ALERT email                       │   │
│  │                                                      │   │
│  │  Flask endpoints: /health  /status  /trigger         │   │
│  └──────────────────────────────────────────────────────┘   │
└─────────────────────────────────────────────────────────────┘

The app is a standard Python Flask application — 256 MB memory, one CF instance, no database required. It uses:

• APScheduler for the daily cron job
• Requests for the UAS and ANS API calls
• Gunicorn as the WSGI server
• XSUAA for JWT-based protection of the REST endpoints

Getting Started

The full deployment takes about 10 minutes:

# 1. Clone the repo
git clone https://github.com/SAP-samples/btp-ai-core-usage-monitor.git
cd btp-ai-core-usage-monitor

# 2. Create required CF services
cf create-service xsuaa application ai-core-monitor-xsuaa -c xs-security.json
cf create-service uas reporting-ga-admin uas-reporting

# 3. Edit manifest.yml — set your subaccount ID, spending limit, SMTP details
# 4. Deploy
cf push

# 5. Set your SMTP password (never in manifest.yml)
cf set-env ai-core-monitor SMTP_PASSWORD 'your-smtp-password'
cf restart ai-core-monitor

# 6. Verify
curl https:///health

The minimum configuration is three environment variables:

env:
  SUBACCOUNT_ID: "your-subaccount-uuid"
  SPENDING_LIMIT: "100"
  NOTIFICATION_EMAIL: "your-team@company.com"

The full set of configurable parameters:

Parameter Default Description

REST API

The app exposes a small REST API for operational use:

Endpoint Description

The /trigger endpoint accepts {"dry_run": true} to fetch and compute usage without sending any notifications — useful for testing your setup.

Try It

The project is open source under Apache 2.0:

👉 github.com/SAP-samples/btp-ai-core-usage-monitor

👉 AI4U project page: ai4u-website.cfapps.eu10-004.hana.ondemand.com/project/btp-ai-core-usage-monitor

If you find it useful, have questions, or want to contribute — issues and pull requests are welcome.