Cost Narrative

Your cloud bill went up. We help you find out why.

CostNarrative is an open source cost investigation workbench that correlates cloud billing anomalies with real engineering events — deploys, autoscaling, config changes, retry storms — and presents ranked hypotheses with confidence scores in plain engineering language.

No more spending 2-4 hours cross-referencing AWS Cost Explorer, GitHub deploy history, Kubernetes events, and AI API usage. CostNarrative builds the narrative for you.

View on GitHub | Documentation

HOW IT WORKS

1. Connect your cloud providers Link your AWS, Azure, or AI API accounts (read-only access). Or simply upload a CSV for instant analysis.

2. CostNarrative detects anomalies Statistical anomaly detection runs automatically per service, comparing hourly costs against a 7-day rolling average.

3. Engineering events are correlated When a cost spike is detected, CostNarrative gathers all engineering events in the surrounding time window (deploys, pod scaling, config changes) and ranks them by correlation score.

4. You get the narrative AI-powered investigation summaries (via your own LLM key) explain what likely happened in plain English — with confidence scores, not certainty claims.

KEY FEATURES

Cost Investigation Workbench

• Unified timeline of cloud costs + engineering events

• Anomaly detection with configurable sensitivity per service

• Causal correlation engine with 5 scoring dimensions

• AI investigation narratives via BYOK LLM

• Markdown export for postmortems and Slack threads

Multi-Cloud Support

• AWS Cost Explorer (IAM role, read-only)

• Azure Cost Management (Service Principal, read-only)

• OpenAI & Anthropic usage tracking

• CSV import for zero-integration analysis

Engineering Context

• GitHub deploy events via webhook

• Kubernetes agent for HPA scaling, OOM kills, pod restarts

• LLM-powered commit analysis for cost-relevant signals

• Config change detection from file path patterns

Optimization Recommendations

• Kill waste: idle resources, orphaned volumes

• Schedule resources: non-prod running 24/7

• Right-size: over-provisioned instances

• Model swap: AI model cost comparison

• Architecture changes: retry reduction, caching

Self-Hosted & Private

• Docker Compose: one-command deployment

• All data stays on your infrastructure

• No telemetry, no phone-home, fully open source (MIT)