Security & privacy

20+ ready-to-use guardrails, end-to-end encryption, per-key whitelists and a full audit trail — security is never an optional module at TokenHawk.

A proxy that sees your AI agents' traffic has to earn that trust. Here is, concretely, what TokenHawk does — and does not do — with your data.

20+ guardrails, 5 categories

emails, phone numbers, GPS coordinates

social security numbers (US, FR, UK), passport numbers

credit cards, IBANs, BIC codes, crypto wallets (BTC, ETH)

AWS, OpenAI, Anthropic, Google, Stripe keys, GitHub/Slack tokens, JWTs, private keys

IPv4/IPv6 addresses, MAC addresses, URLs carrying credentials

Each API key only accepts traffic from the addresses or ranges you declare.

Restrict a key to only the models it needs — a compromised integration can't switch to a more expensive model.

Cut off a key's access in one click, or restrict it to a specific time window.

Every block — whitelist, guardrail, quota — is logged and queryable, ready for your compliance reviews.

Your provider keys are encrypted with AES-256-GCM — never stored in plain text, not even readable by us.

The content of your requests is never used to train a model, nor shared with a third party for commercial purposes.

The text of your requests is not stored in our databases — only usage metadata is (timestamp, model, token count).

Our infrastructure — application servers, database, caches — is hosted in the EU, on AWS.

These guarantees are detailed, with their limits, in our privacy policy.