Live · updated 7 minutes ago · 88 stories in 24h

← Stream
AI & Tech Confirmed First seen

Vitalik Buterin releases privacy-focused zkAPI client with Tor and local LLM rewriting

Vitalik Buterin posted about wrapping zkAPI, Tor, extra cover traffic and local LLM prompt rewriting into an OpenAI-compatible daemon, plus a "paranoid mode" wrapper for his agent harness (pi). He states each conversation gets a separate Tor circuit, requests are masked with random cover traffic, zkAPI prevents identity leaks through payments, and prompts are locally rewritten using a bring-your-own-local-model setup. He notes increased latency, especially when starting a new conversation, and says Tor is far from optimal for privacy, with cover traffic being weaker against stronger adversarie

Why “Confirmed”

  • Published by a primary source: Vitalik Buterin

1 independent source group across 1 post · evidence score 90/100. How labels work

What the sources claim

  1. Buterin wrapped zkAPI, Tor, extra cover traffic and local LLM prompt rewriting into an OpenAI-compatible daemon and made a paranoid mode wrapper for his agent harness pi.

    “Wrapped zkAPI + Tor + extra cover traffic + local LLM prompt rewriting in an open-AI compatible daemon, and made a "paranoid mode" wrapper for my agent harness (pi):”
    Vitalik Buterin
  2. Each conversation is assigned a separate Tor circuit and requests are masked with random cover traffic.

    “Each conversation gets assigned a separate Tor circuit, your requests are masked with random cover traffic”
    Vitalik Buterin
  3. zkAPI ensures identity is not leaked through payments.

    “zkAPI ensures you don't leak your identity through payments”
    Vitalik Buterin
  4. All prompts are locally rewritten using a bring-your-own-local-model setup, with qwen3.8-flash-next being used.

    “all prompts are locally rewritten (bring-your-own-local-model; I'm using qwen3.8-flash-next)”
    Vitalik Buterin
  5. The agent harness is wrapped with a bwrap sandbox and torsocks so everything it does goes through Tor.

    “My agent harness is wrapped with a bwrap sandbox (also experimenting with MicroVM) and wrapped again with torsocks, so everything it does goes through Tor.”
    Vitalik Buterin
  6. There is increased latency, especially when starting a new conversation, and Tor is far from optimal for privacy, with cover traffic weaker against stronger adversaries.

    “you definitely still pay quite a bit in increased latency, especially when starting a new conversation. Also, Tor is far from optimal here for privacy; the cover traffic mechanism is okay against passive adversaries but much weaker if the LLM…”
    Vitalik Buterin

Sources

More in AI & Tech

All AI & Tech stories →

Summary and claims extracted by AI with prompt analyze-story@5 on 2026-10-17 00:00 UTC. Labels are computed from the sources, not by the model.