Free cloud & AI cost tools

33 tools and 267 pages of reference data for the cost decisions that actually move a cloud bill. No login, no email, no upload, every calculation runs in your browser, and the ones that read your billing data have no server to send it to.

CloudType

What are you trying to do?

8 situations these tools were built for, each with the order you would actually work through them and what you need in front of you before you start.

My bill went up and nobody knows why

Diff two months at SKU level to find what moved, then check whether the mover was a vendor price change rather than your own usage.

Have ready: Two billing exports covering equal-length periods, in the same currency.

I need to find savings in AWS

Work down from the changes that carry almost no risk to the ones that are an architecture decision. Storage class and volume type first, idle capacity next, network last.

Have ready: Instance types and counts, volume sizes, bucket size and object count.

I need to plan or right-size Kubernetes

Size the cluster from pod requests rather than node count, see the gap between requested and used, then decide whether a serverless runtime removes it more cheaply than tuning does.

Have ready: Total CPU and memory requested by pods, observed utilisation, node shape and rate.

I need to control BigQuery spend

Find which queries are expensive and why, fix the query shapes that scan more than they need, and only then ask whether committed slots would beat on-demand.

Have ready: A query to analyse, or 30 days of hourly aggregates from INFORMATION_SCHEMA.

AI spend is growing faster than anyone planned

Price the token workload, then price the agent loop around it, which is where the multiple lives.

Have ready: Model, tokens in and out per call, and for agents the steps and fan-out per run.

I am comparing data platforms

Translate one platform's unit of billing into another's, then model each on its own terms, because a credit, a DBU and a slot-hour are not interchangeable.

Have ready: Your credit or DBU rate, warehouse size, query volume and average query length.

I am on Azure or Google Cloud, not AWS

The same cost questions, on the two clouds whose pricing works differently. Azure bills a managed disk by SKU rather than by gigabyte, and Google applies sustained use automatically in a way that a commitment can actually undo.

Have ready: VM size and count, disk sizes actually used, bucket size, and your uptime pattern.

Our billing data is not good enough to act on

Find out how much of the bill you can attribute today, check the export against the FOCUS specification, and get a read on where the practice stands.

Have ready: One billing export. Nothing is uploaded; it is parsed in your browser.

Or explore by cost driver

Categories are organised by cloud, which only helps if you already know which service is responsible. These cut across all of them, so an egress problem finds the NAT gateway, the CDN and the cross-AZ pod traffic tools in one place.

267 pages of dated, sourced reference behind the tools: one page per FOCUS column, per query recipe, per anti-pattern, per model. Free, no login, and written to be linked to.

FOCUS column dictionary

123 pages

Every column in the FOCUS 1.4 billing spec, with its data type, null rules and the requirements a validator checks it against.

BigQuery FinOps query pack

24 pages

Copy-paste INFORMATION_SCHEMA queries for the questions a BigQuery bill raises: cache hit rate, cost by label, daily spend anomalies, slot contention.

BigQuery cost anti-patterns

14 pages

The query shapes that scan more than they need to, each with a worked example of what it costs and the rewrite that fixes it.

Google Cloud SKU dictionary

52 pages

Look up what an opaque SKU on a Google Cloud invoice actually bills for, browsable by service and resource family.

Cloud price change tracker

3 pages

SKU-level price movements per cloud, with an Atom feed so a change reaches you rather than waiting to be found.

LLM price index

51 pages

Per-model input, output, cache and batch rates for every model in the catalog, plus head-to-head comparison pages for the pairs people actually choose between.

Why these are free

Each of these answers a question our platform answers continuously against your live accounts. The tools here work on a snapshot you provide by hand; Finitizer connects read-only and keyless, re-evaluates every day, catches drift, and ships the fix as a Terraform PR or a Jira ticket. If the snapshot version is useful, the continuous version is more so.

See how Finitizer works →