Cloud spend, turned into tiered and auditable action.
Low-criticality changes apply themselves inside a 15-minute rollback window. Medium waits for your approval. High escalates with a business case. Every figure names its basis and the window it was measured over.
Ingest, recommend, act
Three steps, one vocabulary. Nothing moves between them without a record.
Ingest
Cost, usage and inventory from your cloud accounts land in one FOCUS-native model through a read-only role you create and can revoke. Hourly metrics, daily billing exports.
Recommend
Each change is priced in dollars and CO2e against the provider's own price list, with a confidence score, the performance risk and the window it was measured over. Never a bare percentage.
Act
Low auto-remediates and watches itself. Medium runs a runbook a named person approved. High escalates with a business case. Every path has a rollback plan before it starts.
Current versus recommended, side by side
The comparison card is what an engineer reads and what a CFO signs. Same object, one basis line.
- vCPU / RAM
- 8 / 32 GiB
- CPU p95, 14 d
- 31 %
- memory p95
- 38 %
- pricing
- on-demand
- monthly
- $6,240.00
- vCPU / RAM
- 4 / 16 GiB
- CPU p95, estimated
- 62 %
- memory p95, estimated
- 76 %
- pricing
- on-demand
- monthly
- $3,120.00
The human is in charge
Criticality decides who decides. The aperture narrows as risk rises, and the word is always on screen next to the colour.
Auto-remediated
Applied automatically inside a 15-minute rollback window, only above a 92 % confidence gate, only when reversible, never on protected resources. The watcher reverts on any error signal, and the reversal is a record like any other.
Human-approved runbook
An ordered runbook with a full rollback waits for a named approver. Nothing runs until they say so, and the approval is signed with a single-use token that never touches the dashboard.
Business case
Escalates to a sponsor with break-even, risk and the p10–p90 range attached. It stays open until a person closes it.
The 15-minute rollback window
Every auto-applied change is watched for 15 minutes against the metric that justified it. Miss the threshold and it reverts on its own, and you are told either way. The pulse means the window is open and nothing else.
One model, four kinds of cost
A dollar in a datacentre and a dollar in a token budget compare on the same line.
AWS through FOCUS
Compute, storage, network and commitments from CUR 2.0 and FOCUS exports. Azure, GCP and on-prem vCenter follow on the same model.
CO2e next to every dollar
Location-based, with the grid factor and the error range stated. The cheapest region is often not the lowest-carbon one; you see both.
ValueOps and BVPD
Tokens, GPU-hours and inference endpoints priced per unit of work, so a model swap is a costed decision rather than a preference.
Cost Designer
Price an architecture before it exists. Compare shapes, commitments and regions against your own rate card, then export the business case.
The numbers come from your bill, not our blog
We do not publish a savings percentage, because we do not know your bill yet. What we show you on day one is the basis of every figure and the window it was measured over.
- Prices come from the provider's own price list, never a hard-coded rate.
- Forecasts are p10–p90 bands with a confidence label, never a single line.
- Carbon carries its methodology and an uncertainty label.
- The AI explains and drafts; it never computes or alters a figure.
- Every recommendation, approval, action and rollback is an immutable audit record you can export.
Connect one account, read-only, and see the first tiered list.
Nothing is changed until you grant write access and approve a runbook. A 45-minute walkthrough against your own bill is the usual first step.