Quickstart
From an empty workspace to a first governed snapshot, and what to look at once the data lands.
Getting useful output from Flux takes three things: a connection, one collection cycle, and about ten minutes of reading the results.
1. Connect a subscription #
Open Integrations and add the subscriptions you want in scope. Each needs a friendly name and its subscription ID.
Flux authenticates with a managed workload identity — there is no secret to paste or store. The identity needs read access to your resources and cost data:
| Capability | Why it is needed |
|---|---|
| Resource read | Inventory, tags, configuration, and change detection |
| Cost Management read | Actual and amortized cost, daily history |
| Advisor read | Vendor recommendations, cross-checked against Flux findings |
| Metrics read | Utilization for right-sizing and idle detection |
2. Run the first synchronization #
Select Synchronize now. The first run takes longer than later ones because it has no history to build on, and it does several independent things at once:
inventory → every resource in scope, normalized and snapshotted
cost → month-to-date actual and amortized, by subscription and resource
advisor → active vendor recommendations
policy → compliance posture and exemptions
telemetry → utilization summaries per resource
Each source completes independently. If one is throttled or fails, the others still land and the failure is shown rather than hidden — you will not get a silently partial estate.
3. Read the results #
Once the run finishes, work in this order:
- Overview — estate shape, month-to-date cost against the prior period, utilization coverage, and open opportunities.
- Data freshness — which sources completed, which are stale, and what coverage you actually have. Read this before trusting any total.
- Opportunities — the actionable queue, separated from findings that need more evidence.
- Reports — cost summary, allocation, commitments, and the executive summary.
What good looks like after a week #
A single run gives you a picture. A week gives you a baseline, which is when the more valuable features switch on:
- Anomaly detection needs several weeks of matching-weekday history before it reports findings rather than a warming-up state.
- Change detection needs at least two snapshots to produce a diff.
- Right-sizing needs a full observation window of telemetry before it will call a machine oversized.
Next steps #
- Connecting your cloud — scopes, permissions, and multi-subscription estates in detail.
- Cost allocation — the usual first configuration task after data lands.