How Infralyst works
From AWS usage data to Terraform pull requests, inside your existing workflow.
See how Infralyst connects to your cloud, analyzes usage, and prepares Terraform changes that your team reviews and merges.
1. Connect Infralyst to your infrastructure
A one-time setup that usually takes less than an hour.
Connect AWS
Add the Infralyst Terraform module to your infra repo. It creates a read-only IAM role so Infralyst can see your AWS resources and usage, but it never deploys changes in your accounts.
Connect GitHub
Install the Infralyst GitHub App in your GitHub organization and choose which infrastructure repositories it can access. These are the repos where Infralyst will open Terraform pull requests.
Map Terraform state
Link your Terraform states to the right AWS accounts and repos. This keeps every recommendation tied back to the exact code that manages the resource.
Once setup is complete, Infralyst runs in the background and keeps watching for new cost-saving opportunities.
2. How Infralyst finds downsizing opportunities
Multi-stage analysis that is deliberately conservative.
Collect and normalize metrics
Infralyst collects CPU and, where available, memory metrics from CloudWatch for supported resources. We require at least 30 days of usable data before we consider a downsizing recommendation, and we prefer 60 days or more when possible.
Check data quality
We run data-quality checks to make sure there is enough recent and representative data to work with. Resources with very little history or very intermittent usage are skipped until we have a clearer picture.
Analyze CPU and memory
CPU and memory are analyzed separately using averages and high percentiles such as p95 and p99. If CPU or memory is consistently too high for a smaller instance type, that resource is filtered out and no downsizing is suggested.
Look for recurring peaks
We also look at longer-term history, up to around a year at lower granularity, to spot recurring CPU and memory peaks. If we see clear spikes that keep returning, that instance is not treated as a downsizing candidate even if the recent average looks low.
3. Confidence scores
So you can see how strong the signal is for each recommendation.
Based on depth of data
Confidence reflects how much history we have, and whether we see CPU only or CPU plus memory. More days of clean data and both metrics generally mean a higher score.
New workloads stay cautious
Workloads with limited or noisy history receive lower confidence scores or are skipped until we have more data. This helps avoid aggressive changes on brand-new services.
Helps you interpret recommendations
Higher scores mean the metrics are more consistent; lower scores mean we either have less data or see more variability. They're still suggestions, but good candidates for a slower, more manual review.
4. From recommendation to Terraform PR
Infralyst prepares the change. Your team keeps control.
Review recommendations
See a list of potential downsizes for each account, with estimated monthly savings and a confidence score.
Decide what to keep
Use your knowledge of the workload to decide whether a recommendation makes sense for your service or environment. You can ignore or dismiss items that do not fit.
Generate a PR
When you are happy to proceed, click Generate PR. Infralyst creates a branch and Terraform pull request in the right repository with a small, focused diff.
Review, test, and merge
Your team reviews the changes, runs your existing tests and pipelines, and then merges like any other infrastructure change. Infralyst never applies changes directly in AWS.
Optional Slack notifications
If you connect Slack, new savings and PR links can be surfaced directly in your channels so engineers see changes where they already work.
5. Controls and safety rails
Built to fit teams that already treat infrastructure as code.
Read-only access
Infralyst connects via read-only IAM roles and a GitHub App. The only writes we perform are creating branches and pull requests.
No hidden automation
Every change is a Terraform diff in one of your repos. There are no scripts or agents running in your AWS accounts; you review and deploy using your existing pipeline.
Conservative defaults
If data is sparse, noisy, or shows strong recurring spikes, Infralyst does not recommend a downsizing.
Per-resource control
You can dismiss recommendations or exclude specific resources so they are not suggested again.
For more on access and data handling, see Security at Infralyst.
Ready to try Infralyst on your own AWS accounts?
Create a workspace, connect AWS and GitHub, and use your 3 free PR credits on real infrastructure. You review and merge every change.
Start free with 3 PRsNo credit card required · Read-only IAM role · Your team reviews and merges every change