Vitalik's privacy-preserving personal AI: local Qwen orchestrator + zkAPI + Tor, no data leaks
kenziyuliu · x · 2026-10-04
Vitalik Buterin shares a self-experiment: use frontier models on his personal health and travel data for personalized diet and exercise advice, without leaking private info to remote providers.
The setup: a local model (Qwen 3.8 Flash Next) orchestrates and calls powerful remote models as tools for higher-level reasoning and knowledge. Privacy is handled in three necessary layers: the local model writes the queries to avoid leaking identity through writing style; zkAPI hides identity in the payment channel; Tor masks networking/IP. A skill file teaches the local model when and how to construct minimal-data requests. It's a practical template for outsourcing privacy-sensitive reasoning to a local model while renting frontier capability from the cloud.
More from coding & agent
- Turning fuzzy judgments into classifiers with Opus 5.5 agents — no manual labeling needed — ivan_bezdomny · 2026-10-04
- Two keyframes in, full motion graphics out: builder says Opus beats Sonnet at web animation — MicahBerkley · 2026-10-04
- 18-agent framework AIPass found agents' first use of email was filing bug reports on each other — Input-X · 2026-10-04
- AI agent negotiates car lease and mortgage, saving 15-20% by voice — armand_ruiz · 2026-10-04
- Teknium says setting up Hermes with Discord is now less confusing and painful — Teknium · 2026-10-04
- Non-technical founder hunting for a full 'AI product team' stack, cites Garry Tan's gstack — PuzzleheadedMonk3471 · 2026-10-04