Skip to content

Reference architecture

The Architecture page explains the split control-plane / data-plane design in prose. This page shows it. Both diagrams are AWS reference architectures (the EC2 variant runs a single instance in place of the Fargate service; everything else is the same).

The agent runs inside your VPC, reads your data in place, and sends back only findings. Your data never leaves your environment.

How Argus works in your cloud: the agent runs inside your AWS account, reads S3, RDS, DynamoDB, Redshift, IAM and KMS read-only, and sends only findings and metadata to the Argus control plane over outbound HTTPS.

Everything the CloudFormation stack creates (left, all argus-prefixed) and everything the agent reads (right, your existing resources). Reads are the default; writes are a separate, opt-in grant that is off unless you enable remediation.

In-account footprint: the Argus CloudFormation stack (ECS Fargate task, Secrets Manager enrollment token, egress-only security group, CloudWatch logs, Task and Exec roles with ArgusDiscoveryRead and ArgusIdentityRead read policies plus an opt-in ArgusRemediationWrite policy) reading your existing S3, RDS, DynamoDB, Redshift, IAM and KMS resources.

For the exact IAM statements behind the read and opt-in-write policies, see IAM permissions. For what the stack creates per deploy shape, see Deploy overview.

For sharing outside the docs (a security review, a follow-up email) - PNG for a quick drop-in, SVG to scale, PDF for both on printable pages: