FOCOTAI
FOCUS-native · every action reversible

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.

AWS today · Azure, GCP and on-prem nextread-only until you approve write accessevery action in an immutable audit log
Recommendations · demo estate · 14-day window
Rollback window open · 12:41 left
Low · auto-remediates3
Stop dev-app-01 outside business hours
saving$63.63 / mo
carbon−41 kg CO2e / mo
confidence 96 · rollback: start_instances · watching
Delete 4 orphaned snapshots older than 90 days
saving$18.20 / mo
carbon−2 kg CO2e / mo
confidence 99 · 7-day reversible hold
Release 2 unattached elastic IPs
saving$7.30 / mo
confidence 99 · approved by ops@ · 09:14
Medium · needs a human2
Rightsize prod-api-03 m5.2xlarge → m5.xlarge
saving$3,120.00 / mo
carbon−58 kg CO2e / mo
p95 CPU 31 % · perf risk low · runbook 6 steps
Buy 1-year Savings Plan for the steady baseline
saving$1,940.00 / mo
coverage 42 % → 71 % · break-even 4.1 months
High · business case1
Move the analytics cluster to a lower-carbon region
saving$4,480.00 / mo
carbon−1.9 t CO2e / yr
p10–p90 $3.6K–$5.1K · sponsor: CTO · open 6 days
Figures shown are from the FOCOTAI demo estate and are labelled DEMO in the product. Your board is built from your own bill.
How it works

Ingest, recommend, act

Three steps, one vocabulary. Nothing moves between them without a record.

Step 1

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.

AWS CUR 2.0 / FOCUS exports · CloudWatch · on-prem vCenter
Step 2

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.

$ delta CO2e delta confidence
Step 3

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.

Low Medium High
Evidence

Current versus recommended, side by side

The comparison card is what an engineer reads and what a CFO signs. Same object, one basis line.

Current
prod-api-03 · m5.2xlarge
vCPU / RAM
8 / 32 GiB
CPU p95, 14 d
31 %
memory p95
38 %
pricing
on-demand
monthly
$6,240.00
Recommended
m5.xlarge
vCPU / RAM
4 / 16 GiB
CPU p95, estimated
62 %
memory p95, estimated
76 %
pricing
on-demand
monthly
$3,120.00
Monthly saving
$3,120.00
Carbon
−0.7 t CO2e / yr
Confidence
94
Rollback
6 steps
Get runbookApproveSnooze 30 daysAuto-apply (Low only)
Basis: 14-day utilisation to 13 Sep 2026 · AWS public on-demand prices, ap-south-1 · location-based grid factorDemo estate · labelled DEMO in the product
The gate

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.

Low

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.

idle schedules · orphaned volumes · stale snapshots · unattached IPs
Medium

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.

instance family moves · commitment purchases
High

Business case

Escalates to a sponsor with break-even, risk and the p10–p90 range attached. It stays open until a person closes it.

region migration · architecture change · tenancy

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.

Watching · 12:41 left
Scope

One model, four kinds of cost

A dollar in a datacentre and a dollar in a token budget compare on the same line.

CloudLive

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.

CarbonLive

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.

AI tokensNext

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.

Design stageNext

Cost Designer

Price an architecture before it exists. Compare shapes, commitments and regions against your own rate card, then export the business case.

Honesty

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.