🔐 Introducing PrivateInference (PI)
🤔 What:
0 compromise private inference - wholesale or retail - one API key, one funded session, access to a universe of open and closed-weightg models from the leading labs, including OpenAI and Anthropic. No supplier accounts, no subscriptions. ✅
💭 Why:
1. Prompts have become highly sensitive data: unlike search queries, they can reveal reasoning, contracts, medical records, unreleased code, negotiations, and other confidential context. This information remains exposed on third-party servers during processing.
2. “Private inference” gets thrown around loosely: people normally think TEE - open weight models where execution is encrypted on the hardware level. Indeed, TEE can protect memory but a gateway still reads your prompt. Zero-retention policies constrain future use, not present visibility. Privacy is a chain — identity, payment, routing, and execution each need their own boundary, and a transaction is only as private as its weakest link.
⚙️ How:
🌐 Private settlement for tokens on Aztec mainnet (sender, recipient, amount never go onchain)
💳 One session funds many calls across models and inference providers
🔀 Provider routing with per-request zero data retention where supported
📅 When:
🟢 Live soon on the
@aztecnetwork mainnet - settlement + session billing.
Next up: 🪪 conditional zkKYC (prove eligibility, not identity), blind routing, and an attested TEE execution tier with no fallback to a standard route if privacy can't be guaranteed.
Full article:
nitter.net/GalacticaNet/status/21…