Security and Deployment

Pooled and dedicated tenant boundaries

Existing pooled tenants use a shared ECS cluster and shared EFS filesystem, with a per-customer ECS service, EFS access point, task IAM role, and per-customer Secrets Manager secret. Dedicated hosted deployments use a per-customer ECS cluster and encrypted EFS only after the operator provisions a named Terraform placement and the placement-aware backend is released; existing tenants are not automatically migrated. Both modes may use a shared management plane and VPC.

• Task IAM permissions and EFS access-point filesystem policies scope workspace access.
• Dedicated placement does not promise separate physical hardware or a separate VPC.

Encryption, recovery, and image checks

Current hosted EFS is encrypted at rest; ECS task mounts enable encryption in transit, and the filesystem policy denies unencrypted transport.

• The customer-record DynamoDB table has server-side encryption and point-in-time recovery; the metrics table has point-in-time recovery. These settings do not provide point-in-time recovery for EFS workspaces.
• ECR images are scanned on push; scan findings still require review and remediation.

Credentials in per-customer managed secrets

GitHub access uses credentials authorized by the customer. Each hosted customer has a per-customer Secrets Manager secret; DynamoDB stores secret ARN references rather than raw token values.

• Task IAM roles scope access to the resources needed by the customer service. Tasks receive credentials at runtime; applications and tools must handle them securely.
• Grant only the GitHub repository permissions needed for the work, and revoke access when it is no longer needed.

Authenticated, scoped management

Protected customer-management API routes validate Cognito JWTs and enforce admin authorization; customer-facing configuration routes check tenant claims. The public contact form is an intentional unauthenticated exception.

• Keep customer and admin access scoped to their respective identities and review permissions regularly.

Access and review stay scoped

Customer repository permissions and human review remain part of the delivery workflow; an agent's proposed changes are not automatic production deployment.

• Confirm authentication and authorization requirements for the configured management and execution environment.
• Review proposed changes and approve merges or deployments through your own policies.

Your repository remains the system of record

Thaumaton works against customer-owned repositories. Existing pooled workspaces use a customer EFS access point; a dedicated filesystem requires operator-provisioned placement.

• Discuss EFS backup and restore objectives, retention, and offboarding for your deployment separately from DynamoDB point-in-time recovery.

Native self-hosted execution

Run Thaumaton on your own Mac or Linux host, or use the documented Docker deployment. You provide repository access, a Copilot-enabled account, runtime dependencies, and configured credit limits. The host needs network access to GitHub, Copilot, and any package, tool, or repository services used by your workflows.

Docker image prerequisites

The documented Docker entrypoint downloads a Linux/amd64 Copilot binary from private S3 using a configured bucket and key plus AWS credentials. This is not an offline image or a claim of arbitrary container-platform or Kubernetes portability; network requirements depend on the configured workflow.