How TokenScope protects your keys, sessions, and data
TokenScope sits in the path between your developers and Claude. This page explains exactly what we do and do not do with your data.
Architecture and data flow
TokenScope operates as a transparent proxy between your developers and the Anthropic API. Developers authenticate via browser SSO and receive a scoped session token. That session token is used to route requests through the TokenScope proxy — developers never configure or receive the real Anthropic API key.
Developer
Runs tokenscope launcher
TokenScope Launcher
Authenticates via browser · Sets proxy env vars
TokenScope Proxy
Policy check · Budget check · Classification · Logging
Anthropic API
Claude processes request · Returns response
What is persisted
- • Request metadata (timestamps, tokens, cost, model)
- • Classification results (relevance score, risk score)
- • Policy decisions (allowed/blocked, reason)
- • Prompt text excerpt (configurable; can be disabled)
- • User and project attribution
- • Anthropic API key (encrypted at rest — AES-256)
What is transient (not persisted)
- • Full Claude response content
- • Decrypted API-key material exists temporarily within authorized service memory while a request is processed.
- • Developer session token (in-memory during request)
Anthropic API key security
The Anthropic API key is encrypted at rest using AES-256. It is decrypted only within the authorized proxy service when required to forward a request to Anthropic, and is never returned to developer devices.
API key storage
Encrypted in the database using AES-256. Never stored in plaintext.
Key transmission to developers
The real key is never sent to developer machines. Developers hold only a scoped session token.
Decryption scope
Decrypted in-memory within the proxy service, per request, only when forwarding to Anthropic.
Key revocation
Admins can revoke any developer's access instantly without rotating the Anthropic API key.
What data is stored and for how long
TokenScope stores request metadata to power cost tracking, audit trails, and reporting. Full Claude response content is not stored.
Data retention by plan
Free
30 days
Startup
90 days
Team
1 year
Business
2 years
Customers can request data deletion at any time by emailing hello@aitokenscope.com. Enterprise plans support configurable retention windows.
Encryption
Data in transit
TLS 1.2+ enforced on all connections (Railway and Vercel)
API keys at rest
AES-256 via application-layer encryption service
User passwords
Bcrypt hashing — not recoverable
For questions about infrastructure-level storage encryption, contact our team during your security review.
Tenant isolation
Application-level tenant scoping is enforced through tenant-aware authorization and data-access controls. Data access queries are filtered by tenantId at the application layer. These controls are included in our testing and security-review processes.
API-level tenant isolation
tenantId filter applied to all data queries
Database row scoping
Every table includes a tenantId column; application-layer queries are filtered to the requesting tenant.
Cache key isolation
Redis cache keys are prefixed with tenantId
Dedicated per-tenant database
Physical database isolation available on Enterprise — contact us for details
Infrastructure and hosting
Backend — Railway.app
- Postgres database
- Redis cache
- Proxy service
- API service
Region: US (primary)
Frontend — Vercel
- Next.js web application
- Static assets
- Global CDN delivery
Region: Global edge network
Additional data residency regions and dedicated infrastructure options are on the Enterprise roadmap. Contact us to discuss your specific requirements.
Access to customer data
Access to production systems and customer data is limited to essential operations personnel. All production access requires authentication.
For detailed information about access controls, audit logging, and privileged access management as part of a formal security review, please contact our team.
How your data is used
Prompt data used to train AI models
No — customer data is not used to train any models
Prompt data shared with third parties
No — data is not sold or shared with third parties
Aggregated anonymised usage analytics
Yes — used internally for product improvement only
Compliance and certifications
The items below reflect our current status and roadmap. We do not claim certifications or compliance status before they are verified and complete.
Google SSO
Available on all plans
Microsoft Entra ID SSO
Available on Business plan
GDPR — data deletion
Customers can request data deletion at any time
GDPR — data processing agreements
DPA templates available on request — contact hello@aitokenscope.com
SOC 2 Type II
On our roadmap — contact us for current status
SAML / SCIM provisioning
Enterprise plan — contact us for details
For security assessment documentation or to discuss compliance requirements, contact our team.
Incident response
In the event of a confirmed data breach or security incident affecting customer data:
- Affected customers will be notified by email as soon as practicable
- Notification will describe what data was involved and what actions have been taken
- Incident reports will be provided to affected Enterprise customers on request
To report a security vulnerability, email security@aitokenscope.com.
Subprocessors
TokenScope uses the following third-party providers to deliver the service.
For a full list of subprocessors or data processing agreements, email hello@aitokenscope.com.
Need more detail for your security review?
We are happy to answer questions, provide documentation, or walk through our architecture with your security team.
To report a vulnerability: security@aitokenscope.com