Data Architecture & Security

How ASCEND stores, protects, and isolates your data

Overview

ASCEND is a governance and enforcement platform for AI agents and MCP servers. This page describes what data ASCEND stores, where it resides, how it is encrypted, and who can access it.

ASCEND governs what AI agents do—database writes, API calls, financial transactions—not just what they say. Every action flows through risk assessment, policy enforcement, and cryptographic verification before execution. Everything is logged for regulatory compliance.

Key Principle

ASCEND never stores your proprietary data, trade secrets, or the content of your AI agent's outputs. ASCEND stores only governance metadata: what action was requested, the risk assessment, the approval decision, and the audit trail.

Data Classification

Data Category	What's Stored	Classification	Retention
Organizational Data
Organization profile	Company name, slug, subscription tier, status	Internal	Account lifetime
User accounts	Email, role, permissions, organization association	Confidential	Account lifetime + 90 days
API credentials	Organization-scoped API keys, 32-char prefix, encrypted at rest	Restricted	Until rotation
Agent Governance Data
Agent registry	Agent IDs, display names, risk thresholds, operational mode	Internal	Account lifetime
Action records	Every submitted action: type, tool, risk score, status, justification, context	Confidential	Per tier (30 days–7 years)
Approval decisions	Approver identity, timestamp, workflow stage, signed tokens	Confidential	Per tier (30 days–7 years)
Policy rules	Smart rules configuration, workflow templates	Internal	Account lifetime
Audit & Compliance Data
Audit logs	Immutable record of all events: actions, approvals, denials, kill-switch, config changes	Confidential	Per tier (30 days–7 years)
Compliance mappings	NIST SP 800-53 control references, MITRE ATT&CK tactic/technique per action	Internal	With parent record
BYOK key metadata	Customer KMS key ARNs, health status, rotation history—never key material	Restricted	Account lifetime

Storage Infrastructure

Component	Purpose	Encryption	Availability
PostgreSQL RDS	All application data (organizations, users, agents, actions, workflows, audit logs)	AES-256 at rest, TLS 1.3 in transit, BYOK envelope encryption available	Multi-AZ with automated failover
AWS Cognito	User authentication, password management, session tokens, MFA	AWS-managed encryption, no passwords in application database	Regional redundancy (AWS-managed)
AWS Secrets Manager	Platform credentials, database connection strings, API signing keys	AES-256 with automatic rotation	Regional redundancy, least-privilege IAM
AWS SNS / SQS	Kill-switch messaging, agent control commands, notification dispatch	Server-side encryption, per-organization queue isolation	Regional redundancy, sub-second latency
AWS SES	Outbound email for critical alerts, daily digests	TLS in transit	Regional (us-east-2)
ECS Fargate	Stateless compute—no persistent data stored on containers	N/A (stateless)	Auto-scaling, multi-AZ task placement

note

All infrastructure runs in AWS us-east-2 (Ohio). Total messaging infrastructure costs approximately $5/month using AWS-native services.

Tenant Isolation (4-Layer Architecture)

ASCEND enforces tenant isolation at four independent layers (SEC-TENANT-001). Each layer operates independently—a failure at any single layer cannot expose cross-tenant data.

┌──────────────────────────────────────────────────────────────────────────┐
│                         TENANT ISOLATION LAYERS                          │
├──────────────────────────────────────────────────────────────────────────┤
│                                                                          │
│  ┌──────────────────────────────────────────────────────────────────┐   │
│  │  Layer 1: Frontend Routing                                        │   │
│  │  Organization-scoped URL paths (/org/{slug}/)                     │   │
│  └──────────────────────────────────────────────────────────────────┘   │
│                                                                          │
│  ┌──────────────────────────────────────────────────────────────────┐   │
│  │  Layer 2: Backend Middleware                                       │   │
│  │  Every API request validated against authenticated organization_id │   │
│  └──────────────────────────────────────────────────────────────────┘   │
│                                                                          │
│  ┌──────────────────────────────────────────────────────────────────┐   │
│  │  Layer 3: Application Logic                                       │   │
│  │  All queries include organization_id; API keys are org-scoped     │   │
│  └──────────────────────────────────────────────────────────────────┘   │
│                                                                          │
│  ┌──────────────────────────────────────────────────────────────────┐   │
│  │  Layer 4: Database RLS                                            │   │
│  │  PostgreSQL Row-Level Security enforces boundaries at engine level │   │
│  └──────────────────────────────────────────────────────────────────┘   │
│                                                                          │
└──────────────────────────────────────────────────────────────────────────┘

Layer	Component	Mechanism	Failure Mode
1	Frontend Routing	Organization-scoped URL paths (`/org/{slug}/`)	Incorrect slug → 404, no data leakage
2	Backend Middleware	Every API request validated against authenticated user's `organization_id`	Mismatch → 403 Forbidden
3	Application Logic	All queries include `organization_id`; API keys are org-scoped	Missing filter → empty result set
4	Database RLS	PostgreSQL Row-Level Security enforces boundaries at engine level	RLS policy denies → zero rows returned

Defense in Depth

Even if application code contains a bug that omits a tenant filter, the RLS policy at the database layer prevents cross-tenant data access. All 393 protected endpoints are validated against this model.

Access Control

Authentication Methods

Method	Used By	Mechanism	Security Controls
Cognito JWT	Human users (dashboard)	OAuth 2.0 / OIDC with tenant-specific user pools	MFA, forced password change, HttpOnly cookies
API Key	Agent SDKs, gateways	32-char prefix + encrypted secret, org-scoped	RLS enforcement, rotation support

Roles

Role	Permissions	Scope
Super Admin	Full access: agents, users, policies, approvals, BYOK, settings	Own organization only
Admin	Manage agents, approve actions, configure policies, view audit logs	Own organization only
Approver	Review and approve/deny pending actions in assigned workflows	Assigned workflows only
Viewer	Read-only: dashboards, audit logs, agent status	Own organization only

Platform Administration

Platform-admin access is restricted to OW-KAI operational personnel with credentials stored in AWS Secrets Manager under least-privilege IAM policies with mandatory rotation. BYOK customers retain independent encryption key control—even platform-level access cannot decrypt BYOK-protected data without the customer's CMK.

Encryption Architecture

At Rest

Component	Method	Key Management
PostgreSQL RDS	AES-256 via AWS RDS encryption	AWS KMS or customer CMK (BYOK)
Secrets Manager	AES-256 with auto-rotation	AWS KMS default
SNS / SQS	Server-side encryption (SSE)	AWS-managed
API Keys	Encrypted storage, 32-char prefix for lookup without decryption	Application-level

In Transit

All external communications use TLS 1.3 minimum via Application Load Balancer with AWS Certificate Manager. Internal service-to-service communication within VPC uses encrypted channels.

BYOK (Bring Your Own Key)

Enterprise customers supply their own AWS KMS CMK for envelope encryption. ASCEND encrypts data encryption keys (DEKs) with the customer's CMK. Customers retain full control—revoking the CMK immediately renders data unreadable.

Capability	Details
Key Registration	`POST /api/v1/byok/keys` — register CMK ARN
Health Monitoring	Automated 15-minute checks with proactive notifications
Manual Rotation	`POST /api/v1/byok/keys/rotate` — trigger re-encryption
Auto-Rotation Detection	Detects CMK version changes, auto-re-wraps DEKs
Key Revocation	Customer disables CMK → data immediately unreadable

warning

BYOK is designed to fail-secure. If ASCEND cannot reach the customer's CMK, all encryption and decryption operations fail rather than falling back to platform-managed keys. See BYOK Encryption for full setup and troubleshooting.

End-to-End Data Flow

Every agent action follows this 10-step path from submission to audit trail:

┌─────────────┐     ┌───────────────┐     ┌──────────────┐
│  1. Agent    │────▶│  2. TLS 1.3   │────▶│  3. Auth     │
│     SDK      │     │   Transport   │     │    Layer     │
└─────────────┘     └───────────────┘     └──────┬───────┘
                                                  │
                    ┌───────────────┐     ┌───────▼──────┐
                    │  5. Policy    │◀────│  4. Risk     │
                    │    Engine     │     │    Engine    │
                    └───────┬───────┘     └──────────────┘
                            │
               ┌────────────▼────────────┐
               │  6. Action Record       │
               │  (written to database)  │
               └────────────┬────────────┘
                            │
          ┌─────────────────┼─────────────────┐
          │                 │                 │
   ┌──────▼──────┐  ┌──────▼──────┐  ┌──────▼──────┐
   │ 7. Notify   │  │ 8. Approval │  │ 9. Execution│
   │ (Slack,     │  │ (Human-in-  │  │ (Signed     │
   │  Webhook)   │  │  the-Loop)  │  │  token)     │
   └─────────────┘  └─────────────┘  └──────┬──────┘
                                             │
                                      ┌──────▼──────┐
                                      │ 10. Audit   │
                                      │    Trail    │
                                      └─────────────┘

Step	Component	What Happens
1	Agent SDK	Agent invokes SDK before sensitive action
2	TLS Transport	HTTPS/TLS 1.3 to pilot.owkai.app
3	Auth Layer	API key validated, RLS session variable set
4	Risk Engine	CVSS v3.1 + NIST 800-30 + MITRE ATT&CK scoring (0–100)
5	Policy Engine	Smart rules evaluated, routing determined
6	Action Record	Full record written to `agent_actions` table
7	Notification	Dashboard, Slack, webhook dispatched per config
8	Approval	Human reviews in Authorization Center
9	Execution	Signed token generated, agent proceeds or blocks
10	Audit Trail	Immutable record of complete lifecycle

Kill-Switch Architecture

The kill-switch provides immediate, organization-wide agent shutdown capability:

Attribute	Specification
Latency	< 1 second (100–500ms typical)
Path	Dashboard → SNS → SQS (per-org) → SDK → Agent STOPS
Notifications	Simultaneous: Dashboard, Slack, Webhook, Email, SMS
Fail Mode	Fail-closed: unreachable ASCEND = blocked actions

┌──────────┐     ┌──────────┐     ┌──────────────┐     ┌──────────┐
│ Dashboard │────▶│   SNS    │────▶│ SQS (per-org)│────▶│ SDK Poll │
│  (Admin)  │     │  Topic   │     │    Queue     │     │  → STOP  │
└──────────┘     └──────────┘     └──────────────┘     └──────────┘
                       │
          ┌────────────┼────────────┐
          ▼            ▼            ▼
     ┌─────────┐ ┌─────────┐ ┌─────────┐
     │  Slack  │ │ Webhook │ │  Email  │
     └─────────┘ └─────────┘ └─────────┘

Fail-Closed Design

If an agent SDK cannot reach ASCEND, all governed actions are blocked by default. This prevents agents from operating outside governance during network interruptions or outages.

Compliance Alignment

Framework	Capabilities	Key Controls
SOC 2 Type II	Audit trails, RBAC, encryption, change management	CC6.1, CC6.3, CC7.2, CC8.1
PCI-DSS	Data protection, access logging, key management	Req 3, 7, 8, 10
HIPAA	PHI access controls, audit logging, BYOK	§164.312(a)(b)(c)(e)
FedRAMP	AWS GovCloud compatible, NIST 800-53 mappings	AC, AU, IA, SC families
NIST AI RMF	AI risk assessment, governance, HITL enforcement	Govern, Map, Measure, Manage

For detailed compliance control mappings, see Compliance.

Data Retention by Tier

Data Type	Developer	Business	Enterprise
Audit Logs	30 days	1 year	7 years
Action Records	30 days	1 year	7 years
Agent Registry	Account lifetime	Account lifetime	Account lifetime
BYOK Key Metadata	N/A	Available	Available

Security Validation

472 automated tests at 100% pass rate across 12 security layers. Every deployment runs the full security test suite before reaching production.

Security Layer	Tests	Status
Rate Limiting	28	Verified
Prompt Security	62	Verified
Code Analysis	54	Verified
Action Governance	48	Verified
JWT Authentication	41	Verified
API Key Validation	37	Verified
RBAC Authorization	56	Verified
BYOK Encryption	44	Verified
Audit Logging	38	Verified
Input Validation	32	Verified
Secrets Management	18	Verified
Security Headers	14	Verified
Total	472	All Verified

Next Steps

Security Architecture Overview - 12-layer defense-in-depth model
Data Encryption - Encryption implementation details
BYOK Encryption - Bring Your Own Key setup and management
Authentication - JWT, API key, and Cognito authentication
Authorization - RBAC roles and permissions
Compliance - Framework-specific control mappings

Overview​

Data Classification​

Storage Infrastructure​

Tenant Isolation (4-Layer Architecture)​

Access Control​

Authentication Methods​

Roles​

Platform Administration​

Encryption Architecture​

At Rest​

In Transit​

BYOK (Bring Your Own Key)​

End-to-End Data Flow​

Kill-Switch Architecture​

Compliance Alignment​

Data Retention by Tier​

Security Validation​

Next Steps​