Data handling

Never stored

• Prompts, messages, tool definitions, images or any other request content.
• Completions, reasoning or any other model output.
• Your API keys: only a SHA-256 hash, which cannot be turned back into the key.

Request content passes through the gateway in memory on its way to the model and back. The store and metadata request fields, which ask some providers to keep conversations, are removed before a request is forwarded.

Stored

• Your account: email address and sign-in identity (through our authentication provider).
• For each request: time, API key, model, which deployment served it (our own clusters or a named partner), token counts, cost, retail comparison, latency and status. This is what billing and your usage dashboard are built from. Operational logs carry the same metadata and never content.
• Payments: for stablecoin deposits, the public on-chain details.

Who processes your requests

A model runs either on Avenro's own inference clusters or at an outside provider: the company that makes it (OpenAI, Anthropic, Google, xAI or DeepSeek), which processes the request under its own data policy. Each model's description on the models page names who serves it; DeepSeek's API runs on servers in China. The x-avenro-execution-path response header says whether a request was served by Avenro's clusters (self-hosted) or by an outside provider (fallback).

On-chain payments

Stablecoin deposits are public blockchain transactions: the paying address, amount, token and your account's deposit reference are visible to anyone. The reference is a random value and contains no personal information.

This page describes how the service works. The privacy policy is the binding document.