Security at Trappr

Responsible Disclosure

If you discover a security vulnerability in Trappr, we ask that you report it to us responsibly so we can address it before any public disclosure.

Please include in your report:

• A clear description of the vulnerability
• Steps to reproduce or a proof-of-concept
• The potential impact you believe it has
• Any relevant screenshots, request/response logs, or code snippets

We commit to:

• Acknowledging your report within 3 business days
• Providing a timeline estimate for remediation within 10 business days
• Notifying you when the vulnerability has been resolved
• Crediting researchers who report valid issues (with your permission)

We ask that you do not publicly disclose the vulnerability until we have had a reasonable opportunity to investigate and remediate. Please do not access or modify data belonging to other users during research.

Encryption

Data at Rest

All customer data is stored in an encrypted PostgreSQL database. Disk-level encryption is applied to all underlying storage volumes.

Provider API Keys

Third-party provider API keys (OpenAI, Anthropic, Google, Mistral, and others) stored in Trappr are encrypted at rest using AES-256-GCM with a unique initialization vector (IV) per ciphertext. The encryption key is derived from your account's authentication secret using a key derivation function and is never stored in plaintext alongside the encrypted data.

Keys are decrypted in-memory only at the moment of a gateway request, are never logged, and are never written to external storage in unencrypted form.

Data in Transit

All communication between clients and the Trappr platform occurs over TLS 1.2+. API endpoints enforce HTTPS. LLM gateway forwarding requests to upstream providers are also made over TLS.

LLM Response Storage

By default, Trappr does not store raw LLM responses. Only request metadata (timestamps, token counts, model, latency, cost estimates, and DLP event flags) is logged. This design limits exposure of potentially sensitive AI output content.

Data Isolation

Trappr enforces strict per-organization data isolation:

• All database queries include organization-scoped filters. It is architecturally impossible for one organization's API requests to read or modify another organization's data.
• Agent API keys (cht_ak_*) are scoped to a single agent within a single workspace. A key compromise in one agent cannot be used to access data from other agents or workspaces.
• Execution tokens (chtx_*) are single-use, time-bounded, and scoped to a specific execution session.
• Rate limiting is applied per organization to prevent resource exhaustion attacks from affecting other tenants.

Authentication and Access Control

• User authentication is handled by Clerk, a dedicated identity platform with support for multi-factor authentication (MFA), brute-force protection, and session management.
• All API endpoints authenticate via HMAC-signed tokens or Bearer tokens with strict scope validation.
• Dashboard routes require an active authenticated session. No unauthenticated access to customer data is possible.
• Canary token endpoints use HMAC signing to prevent forgery of trigger requests.

Compliance Status

Enterprise customers requiring specific compliance evidence, security questionnaires, or custom DPA terms should contact hello@trappr.io.

Infrastructure

Trappr runs on major cloud infrastructure providers with:

• Private network isolation between services
• Regular automated database backups with point-in-time recovery
• Uptime monitoring and on-call alerting for platform incidents
• Dependency pinning and automated vulnerability scanning in CI/CD

DLP Security Model

Trappr's DLP masking operates as an in-path proxy. Sensitive patterns are detected and replaced with opaque tokens before the request leaves your environment for the LLM provider. The mapping between original values and masked tokens is stored in-memory per request and discarded after the response is unmasked. No PII or sensitive values are written to permanent storage in their original form during DLP processing.

Contact Security

For vulnerability reports: security@trappr.io

For general security questions: hello@trappr.io