Admin, Governance & Security

Aigenzey Cowork is engineered from the ground up for enterprise governance, strict tenant isolation, auditable execution, and cryptographic security.

Organizations, Users & Roles

Cowork enforces a multi-tenant hierarchy with granular permission tiers:

Departments & Access Boundaries

Organizations can segment users into logical departments (e.g. Sales, Engineering, People/HR, Finance, Legal).

• Skill Boundary Enforcement: Skills authored for specific departments are only visible and callable by members of those departments.
• Connector Department Filtering: Enterprise connectors like Salesforce, NetSuite, Workday, and Stripe can be restricted to specific departments (e.g. SALESFORCE_ALLOWED_DEPARTMENTS="sales,revops").

Enterprise Single Sign-On (SSO)

For Business & Enterprise customers, Aigenzey Cowork provides native Single Sign-On (SSO) integration supporting OpenID Connect (OIDC) and SAML-compatible federation with industry-leading Identity Providers (IdPs).

SSO Architecture & Overview

Configuring enterprise SSO enables streamlined, secure employee access through your organization's centralized identity management platform.

• Authorized SSO Callback URI: When creating the app in your IdP console, register:
  https://cowork.aigenzey.com/auth/sso/callback
• Anti-CSRF & State Security: All SSO authorization handshakes use signed cryptographic state tokens generated with OAUTH_STATE_SECRET, expiring automatically after 10 minutes.
• Secret Encryption: Identity Provider client secrets are encrypted at rest using Fernet symmetric encryption and never exposed in browser payloads.

Supported Identity Providers & Setup Guide

Organization Administrators can configure SSO from Admin Panel → Settings → Single Sign-On:

JIT Provisioning & Domain Enforcement

Tailor your organization's authentication policy to match your IT security posture:

• Just-In-Time (JIT) User Auto-Provisioning: When enabled, new employees signing in with your corporate SSO are automatically provisioned with active Cowork accounts, eliminating manual invite bottlenecks.
• Default Department & Role Assignment: Choose the default organizational role (e.g. Member) and pre-assign initial departments (e.g. All Hands, General Engineering) for auto-provisioned users.
• Domain Whitelisting (allowed_domains): Specify authorized email domain suffixes (e.g. acme.com, emea.acme.com). Only accounts matching these domains will be permitted to authenticate through your SSO flow.
• SSO Enforcement (enforce_sso): Mandate SSO as the sole authentication mechanism for your domain. When enabled, password logins and unmanaged third-party sign-ins are blocked for corporate domain users.
• Safe Test Connection Flow: Before turning on strict SSO enforcement, the Admin Panel includes a Test SSO Configuration probe to verify token exchange and profile mapping, safeguarding your team against administrative lockout.

Token Quotas & Usage Metering

To prevent runaway spend and ensure fair allocation across teams, Cowork provides real-time token budgeting:

• Pre-Run Quota Guard: Before starting an agent run, the orchestrator checks the organization's monthly token quota in Redis. If the budget is exhausted, the request fails with a clear 429 Quota Exceeded alert.
• Post-Run Atomic Accounting: When a run concludes, actual model token counts (prompt tokens, cached tokens, reasoning tokens, output tokens) are atomically deducted using Redis pipeline counters and persisted to PostgreSQL.
• Fail-Open Resiliency: If the quota cache encounters an outage, the system fails open so business-critical workflows are not abruptly blocked.

Audit Logging & Observability

Every agent interaction is recorded in the immutable audit repository (Admin Console → Audit Logs):

Security & Fernet Encryption

Aigenzey Cowork follows industry best practices for credential isolation and encryption:

• Fernet Encryption at Rest: All sensitive credentials (OAuth tokens, refresh tokens, database passwords, custom API keys, SSO client secrets) are encrypted using AES-128-CBC with HMAC-SHA256 authenticated encryption via TOKEN_ENCRYPTION_KEY.
• Zero Exposure in Browser: Secret values never round-trip to the frontend application. Admin consoles only display redacted prefixes for verification.
• Anti-CSRF OAuth State: All authorization redirects carry an HMAC-signed state parameter generated with OAUTH_STATE_SECRET, preventing replay and CSRF injection attacks.

Prompt Injection Defense

Because autonomous agents interact with untrusted external content (emails, ticket comments, web pages, and file uploads), Cowork employs a multi-layered defense architecture:

1. Input Boundary Demarcation: External document content and email bodies are wrapped in structured data envelopes, preventing prompt override instructions from being interpreted as system commands.
2. Tool Scope Sandboxing: Dangerous operations (deleting data, sending outbound bulk emails) require explicit confirmation prompts before execution.
3. Runaway Loop Breaker: The orchestrator enforces strict model-call ceilings (ORCHESTRATOR_MAX_LLM_CALLS=75, SUBAGENT_MAX_LLM_CALLS=50), terminating tool loops safely if an agent enters an infinite retry cycle.

Data Retention & Compliance

Cowork provides comprehensive data sovereignty and compliance controls:

• Right-to-be-Forgotten: Users and administrators can permanently purge individual conversations, memory entries, or entire organization accounts from all database tables and caches.
• No Training on Customer Data: Customer inputs, conversation logs, and connected documents are never used for AI model training.
• Self-Hosted / Hybrid Deployment: Enterprise customers can deploy Cowork inside their own private VPC or cloud environment (Google Cloud, AWS, Azure).