How Consize Works

Consize turns cost and usage signals into governed, verifiable optimization actions. Every optimization, whether it's a Kubernetes rightsizing change or cleaning up an idle cloud resource, moves through the same loop:

flowchart LR
    A[Observe] --> B[Analyze]
    B --> C[Recommend]
    C --> D[Review]
    D --> E[Apply]
    E --> F[Verify]
    F -.->|regression detected| A

Observe. Discovery plugins collect resource inventory and usage evidence, for example Kubernetes workload metrics from Prometheus, or cloud billing and utilization data.
Analyze. The recommender profiles real usage (CPU/memory p95/p99 over a rolling window) rather than relying on static assumptions.
Recommend. A proposed change is generated, along with the evidence and reasoning behind it, and optionally enriched with a price estimate.
Review. The policy engine decides what happens next: block, split into smaller steps, require approval, or auto-apply, depending on your configured guardrails. See The Safety Net for the full decision matrix.
Apply. Approved changes are applied gradually. Large changes are broken into small, reversible steps rather than applied all at once, either as a pull request against your IaC repo or as a guarded runtime change.
Verify. Real SLIs (OOM kills, CPU throttling, latency) are monitored after every step. A regression triggers an instant, byte-identical rollback, no manual intervention required.

Components

Consize is built from a small set of focused pieces that map directly onto the loop above:

Component	Role
Discovery	Finds resources and collects the evidence (metrics, billing data) needed to analyze them
Recommender	Turns evidence into a deterministic, reproducible recommendation
Cost enrichment	Attaches a transparent price estimate to a recommendation, when a pricing source is configured
Policy	Evaluates every proposed change against your configured guardrails before it's allowed to proceed
Orchestrator	Applies approved changes step-wise, within configured boundaries
Safety / Verifier	Watches SLIs after each step and triggers an automatic rollback on regression
Store	Persists resources, recommendations, and their audit history. See the Universal Resource Model for how Consize represents infrastructure resources.
Audit	Records what was observed, recommended, approved, applied, and verified, so every action is traceable

Plugin architecture

Discovery, metrics, and pricing sources are pluggable. Consize ships with built-in plugins for Kubernetes, Prometheus, and cloud pricing, and can load additional plugins so the same Observe → Verify loop extends to new resource types and platforms without changing the core engine. See Supported Platforms for what's supported today.

See it end to end

The fastest way to understand the loop is to watch it happen. The interactive sandbox runs entirely on your machine, no cluster, no cloud account, just Docker:

docker run -p 3000:3000 -p 8080:8080 -it ghcr.io/consize-oss/consize-sandbox:latest

Open http://localhost:3000 and watch the Verifier catch an intentional regression on a seeded checkout-api workload and trigger an automatic rollback in real time.

Try the Interactive Sandbox

Next steps

Was this page helpful?

Suggest edits Raise issue