This is what a 14-day read-only pilot returns for your estate — a genuine architecture-cost review, here generated from a fictional demo estate so you can read the shape of the output before connecting anything. Every figure is projected and stays projected until your real bill settles.
AWS estates only. Azure and Google Cloud estates are not evaluated. 30-day window: 16 May 2026 – 14 Jun 2026.
CloudJaeger
Architecture-cost review AWS only
Aurora Robotics (Demo)
Generated 2026-06-15 00:00 UTC
Advisory only — CloudJaeger executes nothing against your cloud.
Executive summary
Architecture health
84 / 100
medium confidence
Total projected savings
$323.50/mo
Projected across 1 recommended action
Not evaluated: rules cover AWS only (5 Azure and Google Cloud resources unchecked).
What to do first
- 1. Rightsize 2 underutilized resource(s)$323.50/mo Projected
- 2. Reduce NAT data-processing spendsaving not measured
- 3. Apply storage lifecycle + snapshot cleanup
Current architecture
Resources reviewed
18
Billed spend
$66,784 · billing 30d
18 resource(s) across the synced estate, grouped account → region → VPC.
Architecture map
Grouped account → region → cluster from your synced inventory. Cost rollups are shown in the interactive review; this print map shows structure only.
Resource-attributed inventory cost
$3,868/mo · known inventory nodes only, not the complete bill
Top billed cost groups
- Amazon Elastic Compute Cloud - Compute$12,710/mo +7% MoM
- Virtual Machines$7,139/mo +7% MoM
- Amazon Relational Database Service$6,759/mo +13% MoM
- Compute Engine$5,051/mo +11% MoM
- Amazon Simple Storage Service$4,046/mo +8% MoM
Recommended target architecture
Conservative — optimize in placeConservative posturelow confidence
Component swaps
2
Node changes
5 changed · 0 added · 0 removed
Projected cost movement
−$323.50/mo Projected
Migration & rollback
Migration risk: LowRollback: Low
Top risks & rollback
2 instance(s) hosting the candidate workload
Containerizing a single, efficient, stable, poorly-understood, or compliance-constrained workload adds operational surface with no density or elasticity payoff — the decision tree refuses those cases.
Mitigation: Keep the original instances (stopped, not terminated) and the pre-migration AMIs/launch templates until the containerized workload has run a full billing cycle; cut traffic back to the instances if regressions appear.
Migration: MediumRollback: High
2 instance(s) on x86 families
This rests on ARM64 compatibility we have NOT confirmed: if any runtime dependency, vendor agent or native binary lacks a working ARM64 build, the move breaks for that instance. A family match alone is not proof the workload runs on ARM.
Mitigation: Relaunch the instance from the original x86 launch template / AMI; the change is a family swap, fully reversible.
Migration: MediumRollback: Low
2 instance(s) with measured CPU headroom
A downsize removes burst headroom: an infrequent spike outside the sample window, a latency-sensitive burst, or burstable-credit behaviour can throttle the smaller size — validate against peak events before applying.
Mitigation: Resize back to the original instance type from the same AMI / launch template; an instance-size change is fully reversible.
Migration: LowRollback: Low
Missing-data caveats
What would firm up these figures. Until then, projections carry the confidence shown above.
- 2 instance(s) on x86 families · blocking — ARM64 dependency-compatibility check per instance (confirm every binary/agent has an ARM64 build); No saving is measured: current and target type prices for each instance’s region and platform, with a priced smaller type for any overlapping rightsize, are required.
- 2 instance(s) hosting the candidate workload · blocking — Memory utilization (p95) for the candidate instances; Request/throughput metrics for the workload; Whether the workload is stateless, adaptable, or stateful; Software licensing terms (container portability); The current deployment model (manual / scripted / container-ready); The team's container/orchestration skill (or willingness to use Fargate)
- 2 instance(s) with measured CPU headroom — Memory and workload fit on the smaller size have not been measured; validate them before applying