Anthouse

Security architecture

Security is not wrapped around Anthouse. It is how Anthouse is built.

Anthouse is the evidence and deployment layer for an organization’s software estate. The control plane is built on AWS in Frankfurt with private data services, separated credentials, KMS-backed encryption, tenant-scoped authorization and releases bound to verified source.

The protection chain

Five boundaries. No single point of trust.

01Identity

Organization accounts and short-lived sessions

02Authorization

Role checks at every protected operation

03Secrets

Envelope encryption with AWS KMS

04Data

Private PostgreSQL in isolated subnets

05Release

Immutable images and commit-bound deployment

Security controls

Strong defaults that are difficult to bypass.

Anthouse protects the complete evidence lifecycle—from the first connector token through the organization record, model-assisted review, governed deployment and export.

Protected01

Tenant isolation

Every organization-scoped operation resolves membership and role before accessing data. Tenant selection never grants membership, and authorization entry points are inventoried and checked in CI.

Protected02

Encrypted connector credentials

Google Workspace refresh tokens and provider credentials use AES-256-GCM envelope encryption. Per-record data keys are wrapped by a dedicated AWS KMS key in Frankfurt.

Protected03

Private data plane

PostgreSQL runs on Amazon RDS in isolated database subnets with no public route. Application access uses separated service credentials, required TLS and pinned RDS trust roots.

Protected04

Restricted workloads

Services run as non-root, unprivileged AWS Fargate containers with Linux capabilities dropped, no public IP, ECS Exec disabled, and tightly scoped task roles.

Protected05

Separated identities

Authentication, runtime, capture, delivery and migration workloads use distinct database identities and secret keys. Ordinary services are explicitly denied the RDS administrator credential.

Protected06

Verified releases

Services deploy from immutable ECR image digests. The release pipeline binds the exact successful commit to its images, applies reviewed infrastructure and verifies the public endpoint.

Protected07

Purpose-bound encryption

Database storage, credentials, provider tokens, service logs, queues, artifacts and release journals use purpose-specific encryption keys and scoped IAM policies.

Protected08

Privacy-minimized evidence

Application review uses bounded activity metadata, capabilities and aggregate counts. Message bodies, prompts and document contents are not required to establish application activity.

Technology stack

One regional control plane, purpose-separated services.

Anthouse owns its authentication boundary and runs its control plane, data services, release infrastructure and operational safeguards on AWS.

ComputeAmazon ECS on Fargate

Isolated service containers in private application subnets.

DatabaseAmazon RDS for PostgreSQL

Private Frankfurt database, TLS verification and separated SQL identities.

SecretsAWS KMS + Secrets Manager

Purpose-specific keys, envelope encryption and scoped workload access.

Storage & deliveryS3 · ECR · SQS · SNS · SES

Encrypted artifacts, immutable images, queues, alerts and authentication email.

NetworkVPC · ALB · WAF · Route 53

Protected HTTPS ingress, private workloads and an isolated database tier.

OperationsCloudWatch + GitHub Actions

Encrypted telemetry and commit-bound, role-scoped releases.

Subprocessors

A short, understandable data path.

Core infrastructure, billing and optional model analysis are separated so organizations can understand exactly which services apply.

ProviderPurpose and dataLocation
CoreAmazon Web ServicesECS/Fargate, RDS PostgreSQL, ECR, S3, KMS, Secrets Manager, CloudWatch, SQS/SNS, SES, networking and WAFOrganization records, account data, encrypted credentials, evidence metadata, artifacts and operational logsFrankfurt, European Union
BillingStripeSubscription and payment processingBilling contact, subscription and payment informationEuropean Union and United States
OptionalOpenAIAdministrator-requested application evidence analysisA bounded evidence dossier chosen for analysis; employee email addresses are excluded from the model contextAccording to the selected OpenAI service

Anthouse gives 30 days’ notice before adding a subprocessor that handles customer data. Contact security@anthouse.io for security and data-processing questions.

Optional device evidence

Find the applications that never appear in an identity audit.

Google Workspace reveals authorized SaaS activity. The optional Anthouse collector adds the local side of the picture: applications, runtimes and network services operating on organization-managed computers, including internally deployed tools and local AI services.

  • Enrolls into one organization with a scoped credential
  • Reports application and local-service evidence over TLS
  • Preserves device and observation provenance
  • Complements Workspace evidence without becoming mandatory
Open a workspace →
managed-device / evidence

Devicedesign-workstation-07Enrolled

PlatformmacOSHealthy


Local serviceOllama · :11434Observed

ApplicationCursorObserved

RuntimeNode.js service · :3000Observed


Evidence remains isolated to the enrolling organization.

Connected discovery fabric

Identity, security, cloud and devices tell one evidence story.

Each connector preserves its source and provenance while Anthouse correlates matching evidence into one application context. Organization administrators control every connection, synchronization and revocation.

Google Workspace

OAuth audit, token grants and managed Chrome usage reveal third-party SaaS activity, active users and granted capabilities.

Microsoft Entra ID

Enterprise applications, service principals, sign-in activity and consent grants expose the Microsoft identity side of the software estate.

Endpoint security

Microsoft Defender, CrowdStrike, SentinelOne and Sophos contribute installed-software, device and security evidence already collected across managed endpoints.

Cloud and code

GitHub, Vercel, Netlify, Cloudflare and connected delivery systems reveal repositories, deployments, providers and technical ownership.

Managed devices

The optional Anthouse collector reports local applications, runtimes and network services that identity and cloud sources cannot see.

See every requested permission and its purpose →

Responsible disclosure

Found something? Tell the people who can fix it.

Email security@anthouse.io. We acknowledge good-faith reports within one business day and keep the reporter updated through resolution.

We do not pursue legal action against good-faith research that avoids other organizations’ data and does not degrade the service.