Security
How Proxygate protects keys and funds
Proxygate sits between independent sellers and agent buyers, so the two things it has to protect are seller credentials and buyer funds. Provider keys stay server-side and are never logged or stored, the proxy exposes neither side to the other, and every paid call settles from a prepaid balance with a signed receipt. Here is what the platform does today.
Provider keys never leave the gateway
- Seller API keys are stored in a managed secret vault (Google Cloud Secret Manager) and decrypted only in gateway memory, at request time, for the duration of the call.
- Keys are never written to logs, responses, the database, or error messages, and they never travel to the buyer. The gateway logs metadata only, never request or response bodies.
- On each forwarded call the gateway injects the seller key server-side, so a buyer consumes the capacity without ever seeing or handling the credential.
A transparent proxy that does not expose either side
- The gateway strips internal routing parameters and upstream credentials before forwarding, and allowlists which response headers pass back, so a buyer never sees a seller IP, headers, or keys.
- Proxygate does not fingerprint upstream traffic, so a provider sees ordinary requests and cannot single out marketplace calls as a class.
- Buyers receive the genuine upstream response, streaming or single-shot, not a rewritten one.
Funds settle from a prepaid balance, with proof
- Buyers fund one prepaid USDC balance on Solana. The gateway meters each call and draws against that balance, so an agent only spends what it uses.
- Balance and escrow mutations run as single atomic operations, so a credit is reserved or released in one step rather than read-then-write.
- Platform settlement transactions are signed through a managed key service (Google Cloud KMS); the signing key is not held on disk in the application.
- Every paid call returns a signed receipt with the exact amount charged and a request id, so spending is auditable and disputable without trusting the platform on its word alone.
Auditing and account protection
- Every key access is recorded to a security audit trail, separate from application data.
- Signup blocks disposable and temporary email domains to reduce throwaway and abuse accounts.
- For spending agents the recommended credential is a scoped, revocable API key tied to a wallet, not the wallet private key, with delegation tokens for time-limited third-party access. The least-privilege option means a compromised agent cannot move funds on-chain.
Certifications
Proxygate is pre-certification: we do not currently hold SOC 2 or ISO 27001 certification. This page describes the controls built into the product today, not a formal audit. We will publish attestations here as they are completed.
Report a vulnerability
Found a security issue? Email info@proxygate.ai with the details and steps to reproduce. We review reports promptly and will work with you on a coordinated disclosure.
Security: frequently asked questions
Provider keys live in a managed secret vault and are decrypted only in gateway memory at request time. They are never written to logs, responses, the database, or error messages, and they are never sent to buyers. The gateway injects the key server-side per call.
No. The transparent proxy strips upstream credentials and internal routing parameters and allowlists response headers, so a buyer receives the genuine response without ever seeing the seller’s key, IP, or headers.
Buyers spend from a prepaid USDC balance on Solana. Balance and escrow changes are atomic, platform settlement is signed through Google Cloud KMS rather than a key on disk, and every paid call returns a signed receipt with the exact amount charged and a request id.
Not at this time. Proxygate is pre-certification. This page describes the security controls in the product today; we will publish formal attestations here as they are completed. To report a vulnerability, email info@proxygate.ai.