Security

The regulation requires a CRM. We’re going further.

The DoD's CMMC regulation, 32 CFR §170.19(c)(2), defines Garde1's regulatory obligation as a single deliverable: a published Customer Responsibility Matrix. Production already runs on FedRAMP Moderate AWS; GovCloud-resident operations and full FedRAMP 20x authorization are on the roadmap. None of it is required by the regulation. All of it is what serious DIB buyers eventually ask for, so we're not waiting to be asked.

Garde1 protects your scope, your System Security Plan (SSP) drafts, and your connector-derived posture — and never CUI. Identity runs on Microsoft Entra under our Official Microsoft Partner status. The full assessor packet — Customer Responsibility Matrix, SSP boilerplate, regulatory citations — lives at /compliance. New to CMMC? Start with the plain-English explainer →

Principles

Zero trust. Least privilege. Privacy by design.

Zero trust

Every request authenticated, every action audited.

Nothing inside Garde1 is implicitly trusted because it's “already in the network.” Every API call is identity-bound (Microsoft Entra OIDC), tenant-scoped at two layers (application-layer filter + database-layer row-level security), short-lived (15-minute access tokens), and recorded in an immutable 365-day audit log. Privileged operations re-prompt for authentication.

Least privilege

The minimum permissions to do the job. Nothing more.

Connectors run read-only by default — write-capable scopes are opt-in per integration for remediation and baseline workflows only, recorded with a per- user, per-workflow consent. ECS tasks use scoped IAM roles; production secrets live in AWS Secrets Manager and never appear in code.

Privacy by design

We don't collect what we don't need.

Garde1 doesn't ask for or accept CUI in the commercial product. Connectors collect configuration and security metadata — never end-user files, mail, or chat content. Evidence is PII-minimized on the way in. The LLM provider operates under a zero-retention contract and is not used to train any model on customer data.

The stack

Five stages. Edge → Identity → Tenancy → Data → Trail.

A request to Garde1 passes through five named control stages before it ever touches your data — and every change leaves an audit record on the way out. Each stage groups the controls that operate there.

1 — Edge

Foreign actors and bots blocked at the network edge, before they reach the application. All inbound traffic terminates at Amazon CloudFront over TLS 1.2_2021 minimum with HTTP/3 (QUIC) negotiation enabled across every distribution (api, app, marketing); viewer protocol policy is redirect-to-https. AWS WAFv2 is attached at the CloudFront tier with managed AWS rule sets — AntiDDoS, Common, Known Bad Inputs — and AWS Bot Control. A U.S.-only geo allowlist drops requests from foreign IP ranges before they cross the CDN. Nothing reaches the application that hasn't already survived all of these.

2 — Identity

Entra OIDC, Garde1-side MFA, NIST-aligned session bounds. Authentication flows through Microsoft Entra External ID over OIDC. Garde1 is an Official Microsoft Partner — a vetted, verified entity in the Microsoft Partner Network with legal, tax, address, and code-of-conduct verification on file and recurring reattestation. SAML and SCIM federate into Entra; Garde1 itself speaks only OIDC. A Garde1-side gate enforces TOTP, WebAuthn passkeys, or single-use backup codes between Entra success and a session token being issued — independent of whether the OIDC provider asserted MFA. Session bounds align with NIST 800-171A AC.L2-3.1.5/.10/.11: access tokens last 15 min, refresh tokens 7 days, SIF max session 8 hours, idle timeout 30 min.

3 — Tenancy

Two-layer tenant isolation, scoped task IAM, secrets in Secrets Manager. Once a session token is valid, every API call is tenant-scoped at two independent layers, and a query has to satisfy both to return data:

  1. Application layer. A typed query builder injects the caller's organization filter from the authenticated session — code paths cannot construct a cross-tenant query.
  2. Database layer. Postgres row-level security is forced on roughly 110 multi-tenant tables — cross-tenant reads are impossible at the engine level even if a bug bypassed the application layer.

ECS tasks use scoped IAM roles; production credentials live in AWS Secrets Manager — never in code, never in committed environment files.

4 — Data

Validated on ingest, PII-minimized at the boundary, encrypted at rest with customer-managed KMS. When connector evidence arrives, the ingest pipeline runs in this order before anything reaches durable storage:

  1. Authenticate the source. Connector run token is verified and tied to the originating organization.
  2. Validate the payload. Schema-checked against the vendor + control mapping. Anything that doesn't match the expected shape is rejected.
  3. Minimize PII. Display names, personal device identifiers, non-administrator user identifiers, and end-user email addresses are dropped or hashed unless a specific control requires them.
  4. Tenant-tag and snapshot. The minimized evidence is stamped with the customer's organization identifier and written behind the same row-level security that gates every other read.

At rest, every store is AES-256. The primary database (Aurora PostgreSQL Serverless v2) is encrypted with a customer-managed KMS key — not the default AWS-managed one — so the encryption boundary is auditable end to end. S3 buckets enforce aws:kms or AES-256 depending on data class; public-access blocks are forced on. A separate application-layer AES-256 key encrypts high- sensitivity values (vendor API tokens, MFA secrets, backup codes) before they reach the database — so even a full database compromise doesn't hand over connector credentials.

5 — Trail

Immutable audit log, PITR backups, public status. Every state-changing action streams through a change-data-capture pipeline into a dedicated audit store with 365-day TTL and point-in-time recovery enabled. Records are immutable for the retention window — there is no UPDATE path on the audit table — and capture the actor, timestamp, prior + new values, IP, user agent, and the related workflow / control / document. Aurora has 35-day automated backups with PITR to any second; the customer documents bucket has S3 versioning on; deletion protection is enabled on both the database and the audit-log table. The status of every monitored endpoint and worker is public, in real time, at status.garde1.com.

FedRAMP building blocks

Built on FedRAMP Moderate-authorized AWS services, in us-east-2.

Every AWS service in Garde1's production path carries FedRAMP Moderate authorization in the AWS US East / West boundary — every one of them, with no exceptions.

This is the distinction buyers care about most. Building blocks — the AWS services Garde1 runs on — are individually FedRAMP Moderate authorized today. The full-system package goes through its own FedRAMP 20x authorization on the roadmap. Until then, the stack gives DIB customers the strongest commercial-region posture available without leaving the FedRAMP service catalog at any layer.

Concretely, every service in Garde1's production AWS stack appears on the AWS FedRAMP services-in-scope list:

Garde1 usesFedRAMP Moderate (East / West)
Aurora PostgreSQLAuthorized
Amazon S3 + GlacierAuthorized
Amazon ECS + ECRAuthorized
Amazon EC2 + VPCAuthorized
Amazon CloudFrontAuthorized
AWS WAF + ShieldAuthorized
Amazon Route 53Authorized
AWS KMSAuthorized
AWS Secrets ManagerAuthorized
IAM + STSAuthorized
Amazon DynamoDBAuthorized
Amazon SQSAuthorized
Amazon SNSAuthorized
Amazon SESAuthorized
Amazon CloudWatch + LogsAuthorized
AWS CloudTrailAuthorized
AWS Certificate ManagerAuthorized
AWS LambdaAuthorized
Amazon EventBridgeAuthorized
The road from here

Today FedRAMP Moderate AWS. Tomorrow GovCloud. After that, FedRAMP 20x.

Each phase below has a defined trigger, a measurable outcome, and a published date. Shipping each one is how we prove our security posture isn't aspirational.

  1. Phase 1Today

    FedRAMP Moderate AWS, US-only region

    The platform runs on FedRAMP Moderate AWS in a US-only region. Your data is encrypted with keys you control, the database physically can't return another customer's rows, every sign-in goes through Microsoft Entra plus a second factor (authenticator app or passkey), vendor API tokens are encrypted with keys that never live in the database, and every action lands in a year-long tamper-evident activity log. Garde1 never asks for or holds CUI — these controls protect your scope, evidence, and connector credentials.

  2. Phase 2Planned

    AWS GovCloud (US) tenancy

    A dedicated aws-us-gov partition deployment for customers whose contracts demand U.S.-person-only operator access and a GovCloud-resident control plane. Identity federates through Microsoft Entra Government — every component sourced from the FedRAMP High catalog.

  3. Phase 3Planned

    FedRAMP 20x authorization

    Full-system authorization through the streamlined FedRAMP 20x process. The GovCloud tenancy from Phase 2 is the assessment boundary; the building blocks are already authorized. This phase wraps the composed platform in its own package.

Responsible disclosure

Found something? Tell us first.

Send reproduction steps using the form below. We acknowledge reports within one business day and commit to a substantive response within five. We will not pursue legal action against good-faith security research that follows reasonable disclosure norms.

One business day · No sales drip