atmos

atmos

Anonymous Ethereum RPC that is fast enough to leave on.

Run it on your machine

curl -fsSL https://atmos.xavio.dev/install.sh | sh

Then add this to MetaMask as a Sepolia network

http://127.0.0.1:8545/11155111

Needs Node 24 or newer. Install guide

Atmos and stock anon-rpc

Atmos does not replace anon-rpc. It runs the same hash-pinned Nym worker through the same browser harness, so the trust model is unchanged. What changes is how calls are scheduled, which routes are trusted, and what happens when one stalls.

Side by side

Stock anon-rpc (Nym worker)Atmos v1
Route choiceThe worker’s default gateway and exit discoveryUniform within the Nym directory’s top performance band; the entry must answer a WebSocket within 5 s
When is a route trusted?When ready resolvesWhen ready resolves and a real call succeeds (2 tries)
Dead route after readyPossible; seen in 2 of 4 workers in one testRejected at boot; the next exit is tried (up to 4)
A call stallsThe caller waits for the worker’s own timeoutAfter 8 s, traffic moves to a warm standby route
10 parallel reads10 connections, opened one after another through the mixnetOne JSON-RPC batch on a warm connection
Provider batch limitsNot trackedRespected (drpc’s free tier: 3 per batch)
Duplicate sendsBoth reach the nodeDeduplicated while in flight
Writes on failovern/aNever retried on another route
InterfaceA standard fetchrequest(url, jsonRpcBody), returning the JSON-RPC response

The same read, both ways

Stock anon-rpc, from its wallet guide:

import { AnonRpcWorker } from "@anon-rpc/browser-harness";

const worker = new AnonRpcWorker({
  address: "0x3aCD9742C68F6d993066Ae25533e09b9e1f60eD0", // Nym worker, mainnet
  preExisting: { rpcProvider: bootstrapProvider },
});

const res = await worker.fetch("https://rpc.mevblocker.io", {
  method: "POST",
  headers: { "content-type": "application/json" },
  body: JSON.stringify({ jsonrpc: "2.0", id: 1, method: "eth_blockNumber", params: [] }),
});
const { result } = await res.json();

Atmos:

import { createAtmosClient } from "./atmos.js";

const atmos = await createAtmosClient({
  bootstrapProvider,
  probeUrl: "https://rpc.mevblocker.io",
});

const { result } = await atmos.request("https://rpc.mevblocker.io", {
  jsonrpc: "2.0", id: 1, method: "eth_blockNumber", params: [],
});

The difference shows up once a wallet does what wallets do: open with a burst of parallel reads, poll, and keep going when a route goes bad.

What Atmos does not change

  • The trust model. You still trust the on-chain worker hash, and the bootstrap provider for one read.
  • What the provider sees. The provider still sees every call’s content. Provider splitting is planned, not built.
  • Timing. v1 does not shape traffic, so sessions can still be re-linked by their timing. The v2 constant-rate schedule addresses that, at a latency cost.
  • Networks. v1 runs on Nym only. The stock harness also runs Nox and Tor workers.

Numbers

The benchmarks page has the measured comparison, run from Cloud Run with headless Chromium.