You know what you spent on AI this month. Your provider dashboard tells you that.
But do you know why Tuesday's spend was 3x higher than Monday? Do you know which model gives you the best quality per dollar for your specific task? Can your ops team see AI costs in the same Grafana dashboard they use for everything else?
Three new features answer those questions.
1. Cost Anomaly Alerts
The problem: Your AI bill crept up 40% last week. Nobody noticed until the monthly invoice. By then, the damage was done — and you still don't know what caused it.
What we built: Automatic spending anomaly detection using z-score analysis against a 14-day rolling baseline. When your spend deviates significantly from the pattern, you get:
- Automatic detection — no manual thresholds to configure. The system learns your normal spending pattern.
- Attribution breakdown — which model, feature, or provider caused the spike. Not just "spending is high" but "GPT-4o spend on your /summarize endpoint jumped 280% on Tuesday."
- Configurable sensitivity — high, medium, low, or off. Tune how aggressive the alerts are.
- Email notifications — get alerted before the bill arrives, not after.
This works for all plans, including free. The alerts are pattern-based, not budget-based — so they catch anomalies even if you haven't set a budget yet.
Privacy-preserving: We analyze dimension labels and spend amounts. Never prompt content.
Competitive context: Only Portkey offers similar automatic spike detection. But Portkey starts at $49/month and requires routing through their gateway. Our detection is passive (SDK-based, no proxy) and starts at €0.
2. Cost-Quality Scatter Plot
The problem: You're using GPT-4o for everything because it "works." But is it the best value? Maybe GPT-4.1 Mini gives 90% of the quality at 20% of the cost for your classification tasks. You'd never know without testing every model — and who has time for that?
What we built: A scatter plot that maps every model's cost against its quality score, with a Pareto efficiency frontier showing the optimal choices:
- X-axis: Cost per task (based on real pricing, including cached/batch discounts)
- Y-axis: Quality score (based on human verdicts from AISpendGuard's Model Lab benchmarks)
- Pareto frontier line: Models on this line are the best quality you can get at that price point. Everything below the line is suboptimal.
- Filter by task type: Classify, summarize, extract, QA, code generation — because the best model depends on what you're doing
The data comes from real benchmarks with human verdicts, not self-reported model scores. Live at /lab/cost-quality — publicly accessible, no signup required.
Why this matters: This is the first product to combine cost tracking and quality benchmarks in a single visualization. No more guessing — pick the cheapest model that meets your quality bar, backed by data.
3. OTLP Export to Grafana, Datadog, and Any OTLP Backend
The problem: Your ops team lives in Grafana. Your engineering team uses Datadog. Nobody wants yet another dashboard to check. AI costs are siloed in the AI tool while everything else is in the observability stack.
What we built: Push AI cost metrics to any OTLP-compatible backend:
- Grafana Cloud — AI cost panels alongside your infrastructure metrics
- Datadog — AI spend in the same dashboards as your APM data
- Any OTLP collector — Jaeger, Honeycomb, New Relic, SigNoz — if it speaks OTLP/HTTP JSON, it works
Setup takes 30 seconds: paste your endpoint URL and auth token in settings. Metrics exported:
| Metric | Description |
|---|---|
ai.cost.total | Total cost in USD |
ai.events.count | Number of API calls |
ai.tokens.input | Input token count |
ai.tokens.output | Output token count |
ai.budget.utilization | Budget % used |
Privacy-first: We export cost metrics and dimensional labels. Never prompts, never completions. Your OTLP backend sees spending data, not content.
Standards-based: OTLP/HTTP JSON — the OpenTelemetry standard. No proprietary format, no vendor lock-in.
This is a Pro feature (€19/month) because it connects AISpendGuard to your existing infrastructure, making it part of your ops workflow rather than a standalone tool.
Together, These Features Change How You See AI Spend
| Before | After |
|---|---|
| Notice cost spike on monthly invoice | Get alerted within hours of anomaly |
| Guess which model is "good enough" | See cost vs quality data for every model |
| Check a separate dashboard for AI costs | See AI costs in Grafana/Datadog alongside everything else |
All three features are live today.
- Free tier: 50,000 events/month, anomaly alerts included, no credit card
- Pro: €19/month for OTLP export, higher limits, and full enforcement features
Covers features: Cost Anomaly Alerts (F30), Cost-Quality Scatter Plot (F34), OTLP Export (F26).