CloudJaeger
Private beta — owner-approved access

The product, module by module

Twelve modules, each labeled with what it knows, how fresh it is, and which kind of dollar it shows — and where a module is thinner than its name sounds, this page says so.

The homepage shows the seven flagship modules — this page explains all twelve, including proof levels and what is still pending.

Illustrative product views — not customer data.

How to read the proof levels

Every module below wears a proof-level badge — the evidence class behind its claims — plus a live or pending chip where one leg of its loop is explicitly open.

owner real data
Backed by our own AWS bill — real synced spend on our own estate, reconciled against the invoice.
owner lab
Proven end-to-end on a real lab environment we stood up ourselves — not a customer's.
sandbox-proven
Live-proven against a vendor sandbox — real API calls, a real object created and read back.
fixture-tested
Validated against fixtures and golden scenarios at every release — not yet against a live external estate.
pending partner
Needs a partner estate to prove — appears only as a pending note, never claimed as achieved.
01

Dashboard command center

One screen answers the Monday question — what changed, why, and what to do next. Deterministic bands walk from the daily-spend trend through the top movers to a ranked next-best-actions strip, and every number carries source, freshness, and confidence on its face. It runs daily on our own estate.

Who uses it: Everyone, daily — the shared Monday screen.

In the loop: See — what changed, why, and what to do next.

Proof level:owner real datalive· runs daily on our own estate
source · freshness · confidence on every number
Cost data · Live · <1h · HighLast 14 days
Daily spend$4,127 /day
anomalyforecast
Top monthly movers · May → Jun
Compute+$18k
Storage+$6k
Kubernetes+$9k
Rightsizing−$12k

Drivers of the month-over-month change

#1Rightsize 12 over-provisioned instances$1,480/mo
#2Delete 46 unattached volumes$310/mo
#3Stop idle dev instances outside working hours$240/mo

Illustrative product view — not customer data.

02

Cost Explorer

Drill spend by service, region, account, and cloud — billed, effective, and list cost perspectives over synced bill rows. Bill-backed, not estimated: synced from your bill, then reconciled against it — within ~1% of CUR on our own estate, window-aligned, and the same check runs on yours in week one.

Who uses it: FinOps + engineering leads.

In the loop: See — drill from the bill to the line that moved.

Proof level:owner real data

Known limit: Multi-cloud, honestly: AWS proven on our own money; Azure and GCP connected end-to-end on real credentials, zero-spend estates — labeled zero, never padded.

bill-backed — synced from your bill
Group byServiceRegionAccountCloud
PerspectiveBilledEffectiveList
Service30-dayΔ
EC2 — Compute$8,120+4%
RDS — Databases$3,480−2%
S3 — Storage$1,940+1%
EKS — Kubernetes$1,210+9%

Illustrative product view — not customer data.

03

Action Center

One ranked, advisory queue across every engine. Each item blends dollar impact, confidence, reversibility, risk, and effort — scored deterministically, with a plain-English rationale. The LLM only writes explanations; it never decides. Dollars are projected, graded against your real bill after you act.

Who uses it: The whole loop — FinOps, engineering owners, and approvers meet here.

In the loop: Decide — the ranked queue where findings become choices.

Proof level:owner real datapending· first bill-graded savings row

Known limit: Savings tiles read “not verified yet” until a real invoice moves — no real row has been bill-graded yet; the pilot produces the first one.

advisory · dollars projected, bill-graded after you act
#1ComputePROJECTED$1,480/mo
Rightsize 12 over-provisioned instances

Evidence: 14-day CPU p95 under 4% · matched to billing line items

Low riskHigh confidenceReversibleLow effort

Ranked #1: highest projected $ at lowest risk — 12 instances under 8% CPU for 30 days

Next: file a Jira ticket for the owning team — advisory only, nothing changes in your cloud.

#2Delete 46 unattached volumes$310/mo
#3Stop idle dev instances outside working hours$240/mo

Illustrative product view — not customer data.

04

Waste Cockpit

Idle, orphaned, overprovisioned, and unattached resources unified over the synced fleet and ranked with reversibility — projected dollars only, with the evidence attached to every row.

Who uses it: Engineering owners — the people who can actually delete things.

In the loop: See → decide — ranked waste feeds the action queue.

Proof level:owner real datapending· populated view — pending partner estate

Known limit: Our own estate shows $0 waste today — and it says $0. The populated view has not yet been proven on a genuinely wasteful estate; we would rather show nothing than fake something.

$0 means $0
Waste by category$4,790/mo flagged
Idle
$2,140/mo
Orphaned
$860/mo
Overprovisioned
$1,480/mo
Unattached
$310/mo
Delete 46 unattached volumesOne-way$310/mo
Rightsize 12 over-provisioned instancesReversible$1,480/mo

Our own estate today: $0 — and it says $0.

Illustrative product view — not customer data.

05

Ownership Cockpit

Every service gets an owner source — configured, inferred, or honestly unknown — and Unowned is first-class, shown at the top of the queue rather than hidden. Spend-by-owner is labeled attributed: overlap allowed, never a fake partition.

Who uses it: Platform teams + FinOps.

In the loop: Decide — an action needs a name attached before it moves.

Proof level:owner real datapending· multi-owner population — pending partner estate

Known limit: Single internal owner today — our dogfood estate resolves to one owner, labeled “internal”. The ladder is real; the population is thin until a pilot estate lands.

owner source shown on every row
Owner source26 services
  1. configured14 services
  2. inferred9 services
  3. unknown3 services
UNOWNED — first-class$3,020/mo

No owner resolved — shown at the top of the queue, never hidden.

Illustrative product view — not customer data.

06

Kubernetes basis ladder

Kubernetes cost visibility with an explicit basis. Four rungs — usage-only, estimated, OpenCost-imported, bill-backed — with machine-checked guardrails so no surface can claim a rung it has not earned. The pipeline is proven end-to-end on a real lab cluster we stood up — not a customer’s.

Who uses it: Platform teams.

In the loop: See — what the cluster really costs, with the basis stated.

Proof level:owner labpending· bill-backed rung — pending partner

Known limit: Our lab cluster is usage-only and says so; the bill-backed rung lights up only with tagged nodes plus CUR coverage.

cost basis on every K8s surface
Kubernetes cost basis4 rungs
  1. 4Bill-backedCUR-backed

    lights up with tagged nodes + CUR coverage

  2. 3OpenCost-importedimported — labeled
  3. 2Estimatedlist price ×24h
  4. 1Usage-onlycurrentno bill → no dollars

    usage-only — and says so

Every cluster is labeled with the rung it's on — a usage-only cluster says so.

Illustrative product view — not customer data.

07

Jira workflow

Jira workflow built in: findings become owned tickets in the tracker the team already lives in. Outbound is live-proven against a sandbox — a real issue was created, linked, and its status read back. Inbound webhooks are wired and fail-closed — 401 without your secret — and remain pending until configured.

Who uses it: Everyone who owns an action.

In the loop: Act — with approval; the ticket carries the action and its status is read back.

Proof level:sandbox-provenpending· inbound live-automation leg

Known limit: No target-level dedupe yet — re-filing the same target would mint a duplicate issue; on the roadmap.

outbound live-proven · inbound fail-closed
#1Rightsize 12 over-provisioned instances
advisory — nothing executes
outbound · live-proven (sandbox)
Jira · sandbox
CJ-3In Progress
inbound · fail-closed · 401 without your secret
Status · back on the recIn Progress — synced back

Illustrative product view — not customer data.

08

Trust & audit

Every action writes an append-only activity row, hash-chained — each row stores sha256(prevHash + row), so edits, deletes, and reorders are catchable. Verification is exposed in the product and documented on /trust; you can check the chain yourself.

Who uses it: Security reviewers + auditors-to-be.

In the loop: Every step, audited — the hash chain runs under the whole loop.

Proof level:owner real datalive· chain verification on /trust
hash-chained · verifiable
Hash-chained audit logevery action logged
#41role.connected#42rec.accepted#43execution.dry_runprevHash 9f2c…hash ← sha256(prevHash + row)#44report.sent
verify it yourself on /trustappend-only

Illustrative product view — not customer data.

09

Architecture Advisor

A deterministic design review before you build: pillar findings with severities and a rough-order-of-magnitude cost figure, validated against 13 golden scenarios at every release.

Who uses it: Architects + platform teams.

In the loop: Decide — review the design before the bill exists.

Proof level:fixture-tested

Known limit: ROM figures are advisory and have never been checked against an external bill — the tag on the number is the claim.

13 golden scenarios · every release
Design review · outputdeterministic
Single-AZ RDS behind a customer-facing API — no failover path
NAT gateway on the hot data path — per-GB processing compounds
No lifecycle policy on log buckets — storage grows unbounded
ROM $2.4k/moadvisorynever checked against an external bill

Illustrative product view — not customer data.

10

Topology & inventory

An auto-generated map of the synced estate — resources by account, region, and VPC with inferred relationship edges and drift between snapshots, exportable as JSON or Mermaid. 72 real resources on our own estate, refreshed nightly.

Who uses it: Architects + platform teams.

In the loop: See — where spend lives in the architecture.

Proof level:owner real datalive· nightly refresh
refreshed nightly · export JSON/Mermaid
Estate topologysynced · refreshed nightly
+2 since last snapshotVPC · productionEIPunownedALBEC2EBSEBSRDS
normal resource changed since last snapshot unowned

Generated from synced inventory — not a hand-drawn diagram.

Illustrative product view — not customer data.

11

Service / unit economics

Two lenses, one module: service economics — cost stops being a blob and lands on services with an SLO grade — and unit economics — cost per request from a real denominator. Denominators are ours today, yours in the pilot — never a made-up ratio. The same lens covers AI cost: token-level spend by feature, tenant, and model, with margin per feature once you attribute revenue.

Who uses it: Engineering + finance together — one denominator both sides read.

In the loop: See → verify — unit cost is the number both sides watch move.

Proof level:owner real data

Known limit: Denominators are partial today — a handful of instrumented events on our own services; the tile says “partial” or shows a dash rather than inventing a number.

attributed · overlap-labeled
Service economicsper service · 30 days
Attributed
$40.38
bill-attributed spend
Cost / request
$0.0031
denominator: partial
SLO
B+
graded from targets
Requests
—
partial — and says so

Illustrative product view — not customer data.

12

Commitments dry-run

Rehearse RI and Savings Plan decisions end-to-end before a euro is at stake: real coverage and recommendations — proven on our own estate — with dry-run executions that always end dry_run_ok and never move money.

Who uses it: Founders/CFOs + FinOps.

In the loop: Decide — rehearsed, never live.

Proof level:fixture-tested

Known limit: Live purchases are hard-blocked today — arming execution requires your written mandate with caps and a named approver on every run.

dry-run default · live hard-blocked
Dry-run history Execution OFF by default
RunPlanResultMoved
Jun 30 · 09:121yr RI · $1,800 capdry_run_ok$0 moved
Jun 21 · 14:471yr SP · $1,200 capdry_run_ok$0 moved
Jun 14 · 11:031yr RI · $600 capdry_run_ok$0 moved

Illustrative product view — not customer data.

Deep dives

Three details that decide whether you can trust the modules above — the scoring anatomy, the exit door, and one path we ship honestly unfinished.

Anatomy of a recommendation

Every item in the queue carries four deterministic scores — the same four chips you see on the expanded row above:

  • RiskHow likely the action is to disrupt a running workload — scored from the evidence, shown in the open.
  • ConfidenceHow strong the evidence is — e.g. a 14-day CPU p95 matched to billing line items, not a single sample.
  • ReversibilityReversible or one-way — so you can sort by “safe and easy” instead of just “big number”.
  • EffortHow much work acting takes, so quick wins rank ahead of quarter-long projects at equal dollars.

The dollar figure wears a PROJECTED badge until your bill moves. Engines are deterministic; the LLM only writes explanations, never decides.

Exports & FOCUS

Your data leaves in open formats — CSV, JSON, and FOCUS, the FinOps Foundation’s open format. A sample of the mapping:

CloudJaeger fieldFOCUS column
Billed costBilledCost
Effective costEffectiveCost
ServiceServiceName
Charge periodChargePeriodStart / ChargePeriodEnd
CSVJSONFOCUScoverage and validation shown on export

Anomaly stream ingest

NOT LIVE-FED

A near-real-time ingest path into the velocity-anomaly engine — built, unit-tested, and fail-closed: malformed input is rejected, and cold-start baselines are provisional and labeled.

metric streamvelocity baselinebreach flag

No live metric stream has ever fed it — and we say so here rather than let you find out in week two. It stays off this page’s claims until one does.

See the output before you connect anything.

A full sample architecture-cost review — health, ranked findings, target architecture, risks, and missing-data caveats — generated from a deterministic demo estate. Every figure projected; not customer data.

View the sample report →

Explore by need

See every module on your own bill.

Read-only role, first findings in the first meeting — on our estate it took sixteen seconds.

Request accessSee pricing