Security Practices | Sinapsis AI

This page describes the security practices Sinapsis AI currently implements to protect data and workflows. Controls can vary by deployment, configuration, and feature path; this page is not a warranty or a security certification.

1. Encryption in Transit

TLS is used for browser-to-server communication, API requests, and configured connections to infrastructure and third-party providers. Production deployments are configured to redirect or reject unencrypted HTTP where the edge configuration supports it.

2. Encryption at Rest

Sensitive data stored by the application is protected according to its type:

API keys — Hashes are stored when the key only needs to be verified; plaintext keys are not recoverable from those hashes
SSO and integration credentials — Encrypted before storage using application encryption keys
Deployment and inference credentials — Encrypted before storage where the deployment path persists them
Passwords — Bcrypt hashes with per-user salts
TOTP secrets — Encrypted before storage

Database backups and disk volumes use the encryption controls provided by the selected infrastructure. Self-hosted operators are responsible for configuring those controls.

3. Tenant Isolation

Sinapsis AI is a multi-tenant platform. Organization-scoped data is protected through:

Organization-scoped queries and access checks
Membership and role validation for organization resources
Organization-scoped API keys and deployment permissions
Workspace boundaries where enabled
Audited and restricted platform-operator access paths

Tenant isolation is a logical application and database control. It is not a claim that every deployment uses a separate physical database or network.

4. Authentication and Access Control

We implement multiple layers of authentication and authorization:

JWT tokens stored in HttpOnly, Secure where enabled, SameSite=Lax cookies
CSRF protection via double-submit cookies and constant-time comparison
Two-factor authentication (TOTP), optional per user
API-key authentication for programmatic access, with keys shown only once at creation when applicable
Role-based access control with Owner, Admin, Member, and platform-operator permissions
Social login through configured OAuth providers with PKCE where supported
Session revocation on logout, credential events, and account controls

5. API Keys and Connection Secrets

API keys for programmatic access are hashed before storage, scoped to an organization, identified by a non-secret prefix, and revocable through the platform where supported.

External software connection credentials, OAuth tokens, SSO secrets, and persisted deployment credentials are encrypted at rest. The plan and connection configuration determine whether credentials are managed by Sinapsis or supplied by the organization. The applicable external provider controls provider-side retention.

6. Audit Logging

Security-relevant and administrative actions are recorded in a partitioned audit log, including authentication and session events, membership and role changes, API-key events, organization settings changes, data export and deletion requests, compliance decisions, and other protected operations.

Audit records may include the actor, action, timestamp, IP address, user agent, request ID, organization, and resource metadata. The current default retention is 365 days, configurable per organization where supported; legal, security, and operational requirements may require longer preservation. Organization administrators can view available audit logs through Settings > Security & Data > Audit Logs.

7. Rate Limiting and Abuse Prevention

Redis-backed rate limiting and plan-based quotas protect authentication, password recovery, two-factor verification, AI, API-key, export, deployment, and other sensitive or resource-intensive endpoints. Exact thresholds depend on the endpoint, plan, environment, and operational configuration and may change as abuse patterns and capacity change.

8. Network Security

SSRF protection: User-provided URLs are validated against private IP ranges, cloud metadata endpoints, and dangerous ports. DNS pinning helps prevent rebinding attacks where applicable.
IP allowlists: Supported organization and deployment paths can restrict access to configured IP ranges.
Security headers: Production edge responses use headers such as CSP, X-Frame-Options, X-Content-Type-Options, and Referrer-Policy where configured.
CORS: Production API origins are allowlisted; wildcard origins are not used for authenticated production access.
Network segmentation: Sandbox services use dedicated networks separate from the main application paths.

9. Backups and Recovery

Backup, restore, and retention behavior depends on the deployment. Cloud operators manage the platform backup process and recovery controls; self-hosted operators are responsible for backup frequency, encryption, restore testing, and disaster recovery.

10. PII Masking

When enabled for an AI route, the PII masking layer redacts supported identifiers before prompts are sent to external model providers. Current patterns include email addresses, Social Security numbers, credit-card numbers, phone numbers, and IP addresses.

Masking is best-effort and route-dependent. It does not replace your obligation to select an appropriate provider, configure retention, and avoid sending data that your organization is not authorized to process.

11. Code Execution Sandbox

User-submitted Python code from workflow code nodes and agent tools runs in an ephemeral, container-isolated sandbox when the sandbox is enabled:

Network access is disabled and the execution root filesystem is read-only
A shared AST policy blocks dangerous modules, dynamic execution, and Python introspection paths
Seccomp, dropped capabilities, no-new-privileges, and process limits are applied to execution containers
CPU, memory, temporary filesystem, concurrency, and execution-time limits are enforced
Execution timeout defaults to 10 seconds in the current beta sandbox configuration
Input and output are capped at a configurable 1 MB default

The current boundary is a hardened container. Sinapsis does not describe this sandbox as VM or microVM isolation.

Optional product analytics runs behind the analytics consent gate. Umami receives configured event and page-view data, while OpenReplay may provide private session telemetry when configured. Analytics can be rejected or withdrawn through the cookie settings. Analytics failures do not block product use.

13. Vulnerability Reporting

If you discover a security vulnerability, please report it responsibly to [email protected]. We will acknowledge receipt within 48 hours where practicable and work to resolve confirmed vulnerabilities promptly.

14. Infrastructure and AI Provider Configuration

For the current cloud-hosted service, primary platform infrastructure is hosted in the European Union, currently in the Netherlands. For self-hosted deployments, the customer controls infrastructure decisions.

Sinapsis AI is model-agnostic:

Models and approved provider routes are available according to the plan and deployment configuration
Provider training, retention, logging, security, and residency behavior depend on the provider and route you select
Local or self-hosted inference remains inside the selected infrastructure when configured that way
Inference data is sent only to providers or infrastructure configured for the applicable workflow, evaluation, or training job

Review each provider's security, privacy, data-processing, and retention terms before routing sensitive workloads.

15. What We Do Not Claim

In the interest of transparency:

We are not SOC 2 Type II certified at this time
We do not offer a guaranteed SLA for free plans
We do not claim that the platform has regular third-party penetration testing on a fixed cadence
End-to-end encryption of AI prompts is not available; prompts are processed at the application layer before routing to providers
The Compliance module does not provide legal advice, certification, HIPAA attestation, or EU AI Act conformity assessment

For questions about our security practices, contact us at [email protected].