zombiecost

Find the zombie resources in your AWS account: things that should be dead but still bill you every month.

Find idle and forgotten AWS resources and see what they cost you every month.

One read-only scan. One table. A number at the bottom you can act on today.

What it finds

Prices are rough us-east-1 on-demand numbers. The goal is order of magnitude, not an invoice.

Checks that depend on usage history (ec2-idle, rds-idle, nat-idle, lb-unused) ignore resources younger than 3 days, so a freshly launched instance is never reported as idle. s3-stale can see writes but not reads; it says so in every finding, because reads are invisible without paid request metrics.

Install

pip install zombiecost

Or from a clone:

python3 -m venv .venv && source .venv/bin/activate
pip install -e ".[dev]"

Run

zombiecost                                  # every enabled region, default profile
zombiecost --profile prod --regions eu-west-1,us-east-1
zombiecost --days 30 --json report.json     # longer lookback, machine-readable output
zombiecost --redact                         # placeholders instead of IDs, names and IPs: safe to paste publicly
zombiecost --skip s3-stale                  # never list object keys
zombiecost --only ebs-unattached,eip-unused # just the fast, metadata-only checks
zombiecost --list-checks

Privacy and security

Nothing leaves your machine except AWS API calls made with your own credentials. There is no telemetry, no update check, no call to zombiecost.com or any other service. You can verify this: the package has exactly one runtime dependency that talks to a network, boto3, and every call it makes is a Describe, List or Get. There is no code path that creates, modifies or deletes anything.

What the tool reads, per check:

s3-stale is the only check that touches anything object-level. It lists keys to find the newest modification time and discards them; keys are never stored or printed. If you would rather it did not list keys at all, run with --skip s3-stale and drop s3:ListBucket from the policy below.

The report is yours. Plain output and --json contain real resource IDs, names, tags and IPs, because that is what you need to go fix things. If you want to share a report, use --redact: IDs, names, IPs and the account number become stable placeholders and the per-finding details block is dropped entirely.

Reporting a vulnerability: see SECURITY.md.

Permissions

The scan is read-only. Attach ReadOnlyAccess, or this minimal policy:

{
  "Version": "2012-10-17",
  "Statement": [{
    "Effect": "Allow",
    "Action": [
      "sts:GetCallerIdentity",
      "ec2:DescribeRegions",
      "ec2:DescribeVolumes",
      "ec2:DescribeAddresses",
      "ec2:DescribeSnapshots",
      "ec2:DescribeImages",
      "ec2:DescribeInstances",
      "ec2:DescribeNatGateways",
      "rds:DescribeDBInstances",
      "elasticloadbalancing:DescribeLoadBalancers",
      "elasticloadbalancing:DescribeTargetGroups",
      "elasticloadbalancing:DescribeTargetHealth",
      "s3:ListAllMyBuckets",
      "s3:GetBucketLocation",
      "s3:ListBucket",
      "cloudwatch:GetMetricStatistics",
      "cloudwatch:ListMetrics"
    ],
    "Resource": "*"
  }]
}

Tests

pytest

Tests run against moto, no AWS account needed.

Roadmap

• Unused KMS keys, Secrets Manager secrets and VPC endpoints (small, fixed monthly charges that add up)
• Stopped instances still paying for their volumes
• Old AMIs and the snapshots behind them
• Hosted version: cross-account role, weekly scans, Slack alerts