Security & data handling

How TokenX Gateway treats your requests, where they go, and how the platform itself is secured.

Data handling

  • No training. Your prompts and completions are never used to train or fine-tune models — by TokenX or by the upstream providers, whose enterprise terms we operate under.
  • No content retention by default. Request and response bodies pass through in memory and are not written to disk or logs. Optional request logging for debugging can be enabled per key and is off unless you ask.
  • Metadata only. We store timestamp, model, token counts, latency, HTTP status, key id and your optional user field for billing and support, for 90 days.
  • Encryption. TLS 1.2+ on every hop: client to gateway, gateway to provider.

Providers

Inference is performed exclusively by enterprise cloud services under their respective enterprise data terms:

  • AWS Bedrock — Anthropic Claude, Meta Llama, DeepSeek, Qwen, Amazon Nova. AWS does not store prompts or use them for model training; Anthropic does not receive customer content.
  • Azure AI Foundry — OpenAI GPT models. Azure OpenAI data is not shared with OpenAI and not used for training.
  • Google Vertex AI — Gemini. Vertex AI does not use customer data to train models.

Default routing uses US regions for widest model availability. The au tier restricts inference to AWS ap-southeast-2 and Azure australiaeast.

Infrastructure

  • Edge and TLS termination on Cloudflare; origin not directly reachable from the internet.
  • Cloud accounts are dedicated to TokenX, with MFA on all root and administrator identities and no long-lived root credentials.
  • Workloads call providers with short-lived, role-based credentials (AWS STS roles, Azure managed identity / scoped keys, GCP service accounts with least-privilege IAM); no shared or personal credentials in production.
  • Provider credentials are held in a secrets store, never in source code or configuration files, and are rotated on a schedule.
  • Audit logging (CloudTrail and equivalents) is enabled on every cloud account.

Access control

  • One named identity per person; least-privilege permissions; MFA everywhere.
  • Production access is limited to the engineers operating the platform and is logged.

Your API keys

  • Keys are stored hashed; we can show a key only at creation time.
  • Each key has its own rate limit, spend cap and allowed-model list; rotate or revoke instantly from the dashboard or by emailing us.
  • If a key is leaked, revoke it immediately and tell us — we will review usage with you.

Compliance

TokenX Pty Ltd is an Australian company (ABN 29 695 969 414) subject to the Privacy Act 1988 (Cth). A Data Processing Agreement is available for Business and Enterprise customers. We are happy to complete security questionnaires and to walk through our controls with your security team.

Reporting a vulnerability

Email [email protected] with details. We acknowledge within one business day and will keep you informed through remediation.