When building a system for autonomous machine-to-machine commerce, security is the primary design constraint. If a database is breached or a private key is leaked, automated scripts can drain resources in milliseconds. Agentix addresses this challenge by adopting a strict zero-trust, non-custodial architecture.
By design, the Agentix frontend and routing servers do not hold user private keys, do not store unencrypted files, and do not process transactions off-chain. Everything is client-side and verified by smart contracts.
Client-Side AES-GCM-256 Encryption
When a vendor uploads a digital product (such as a database, an SDK wrapper, or an AI model weight file) to the marketplace, the file is encrypted locally in their browser or SDK runtime before upload. The unencrypted file never touches our network. The encrypted blob is stored on decentralized storage networks, and the decryption key is managed strictly through our x402 key exchange protocol.
This client-side security architecture ensures that Agentix servers are never a high-value target for hackers. Even if our servers are compromised, all user files remain encrypted, and the attacker cannot access the underlying data or keys.
Local Secret Management
The Agentix SDK acts as a local library inside your agent's runtime. It handles transaction signing and key derivation locally using standard cryptography libraries. You do not need to share your private keys with Agentix's API. This ensures that even in the unlikely event that the Agentix website interface is compromised, your funds, keys, and data remains completely secure.
Developers can run their agents in secure environments, using environment variables or hardware security modules (HSM) to keep their private keys protected. The SDK interacts with these local stores to sign transaction payloads without exposing secrets to the outside world.
Safeguarding against Prompt Injection
As AI models read untrusted web pages, they are vulnerable to prompt injection attacks that could trick them into calling the payment tool to transfer funds to malicious wallets. Agentix protects against this by enforcing strict local budget guardrails.
The developer can define transaction rules, such as maximum payment amounts per request, destination address allowlists, and daily spend limits. These filters are executed at the local SDK level, completely bypassable by the LLM logic, ensuring that even a compromised model cannot drain the agent's wallet.
