v0.9 · multi-month assessment

Find AWS & Azure waste in under a minute.

Drop 3–12 monthly billing exports. Get a board-ready report with Reserved Instance / Savings Plan recommendations, Terraform + CLI remediation, and AI narration — every number cited to a specific line item.

AWS CURAWS PDF invoiceAzure billing Account IDs hashed
Drop 3 to 12 monthly bills here
CSV · CSV.gz · Parquet · PDF · up to 200 MB total
or
Auto-detects AWS CUR, AWS PDF invoice, and Azure Cost Management export. Account IDs and subscription IDs hashed before analysis. No data leaves your server.

Cited findings, not vibes

Every dollar claim traces to specific bill line items. The deterministic engine emits the numbers; the LLM only writes prose around them and is rejected if it can't cite.

Remediation, not just diagnosis

Each finding ships with a Terraform stanza and AWS CLI command you can paste. Investigation steps come before destructive fixes — never the other way around.

Agentless, self-hosted

No cross-account IAM. No installed collectors. Caddy ingress, API-key gated engine, SQLite persistence. Lives on a $5 VPS or your existing platform.

AWS & Azure on day one

Auto-detects CUR vs Azure Cost Management export. Same pipeline, same citations. Optimizer correctly skips clouds it doesn't model — no fabricated recommendations.

Pipeline

The math runs in Python. The LLM only writes.

Four stages, each owned by a separate module. The narrative layer receives findings with explicit IDs and is rejected if it omits citations. No invented numbers ever leave the engine.

01

Parse + anonymize

Auto-detect AWS CUR vs Azure export. Hash account/subscription IDs and drop PII tags at parse time — anonymization happens inside the parser, not as a downstream step.

02

Deterministic analysis

Service/region/month breakdown, anomaly detection on M-o-M deltas, idle-NAT inference from billing shape, untagged-spend governance signals.

03

LP-solved commitments

1-year no-upfront Compute SP sized to the minimum monthly OnDemand baseline per (family, region) via a pulp linear program. Conservative, citable.

04

Cited narrative + remediation

Claude writes the prose; the validator rejects any paragraph without a finding citation. Each finding gets Terraform + CLI artifacts attached deterministically from metadata.

Methodology

Bills tell you a lot. They don't tell you everything.

We're explicit about what billing-derived analysis can and cannot do. If you want IAM audits or CIS coverage, those need a read-only role — different shape of product, different trust model.

What's in scope

  • Service / region / month cost breakdown
  • Month-over-month anomaly detection
  • 1yr no-upfront Compute SP optimization (AWS)
  • Idle NAT-gateway detection from billing shape
  • Tagging hygiene / untagged spend share
  • Reserved Instance ladder modeling

What's out of scope (and why)

  • CIS / NIST / HIPAA / SOC 2 mapping — bills don't carry config
  • Right-sizing without utilization data — CPU/mem isn't in a CUR
  • Public S3 bucket detection — requires API access
  • IAM / MFA / encryption checks — requires API access
  • Azure RI / Savings Plan optimization — coming in v3

Drop a bill. Get a board-ready report in under a minute.

No credentials. No cross-account IAM. No agents. The whole pipeline runs on your own server.