惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

G
Google Developers Blog
阮一峰的网络日志
阮一峰的网络日志
博客园 - 聂微东
F
Fortinet All Blogs
H
Help Net Security
让小产品的独立变现更简单 - ezindie.com
让小产品的独立变现更简单 - ezindie.com
D
DataBreaches.Net
MyScale Blog
MyScale Blog
B
Blog
I
InfoQ
钛媒体:引领未来商业与生活新知
钛媒体:引领未来商业与生活新知
GbyAI
GbyAI
Google DeepMind News
Google DeepMind News
IT之家
IT之家
The GitHub Blog
The GitHub Blog
有赞技术团队
有赞技术团队
博客园_首页
L
LangChain Blog
V
V2EX
OSCHINA 社区最新新闻
OSCHINA 社区最新新闻
T
The Blog of Author Tim Ferriss
Cyber Security Advisories - MS-ISAC
Cyber Security Advisories - MS-ISAC
Microsoft Azure Blog
Microsoft Azure Blog
博客园 - Franky

Hacker News - Newest: "AI"

AI can't read an investor deck AI as an attorney? Student uses ChatGPT, Gemini to sue UW over alleged racial discrimination Hacking MCP Servers in AI Systems – The Rug Pull: Tool Changes After Approval GitHub - MeepCastana/KubeezCut: Free Web based video editor Can AI judge journalism? A Thiel-backed startup says yes, even if it risks chilling whistleblowers Coming soon: 10 Things That Matter in AI Right Now DARPA built an AI to fact-check enemy weapons claims What explains heterogeneity in AI adoption? When AI Meets Muscle: Context-Aware Electrical Stimulation Promises a New Way to Guide Human Movements - Department of Computer Science AI Changed How We Build. It Did Not Change What Matters. Linux rules on using AI-generated code - Copilot is OK, but humans must take 'full responsibility for the… Meta spins up AI version of Mark Zuckerberg to engage with employees Code Mode: Let Your AI Write Programs, Not Just Call Tools | TanStack Blog GitHub - Delavalom/graft: Go framework for building AI agents. Type-safe tools, multi-provider (OpenAI, Anthropic, Gemini, Bedrock), zero vendor SDKs. India's TCS tops estimates, says new AI models did not dent services demand Gen Z's fading AI hype Strong feeling: we are in a folded AI reality GitHub - machinarii/total-recall-catalog: A reference catalog of latest knowledge retrieval, memory & RAG systems GitHub - mensfeld/code-on-incus: Give each AI agent its own isolated machine with root, Docker, and systemd. Active defense detects and stops threats automatically.. Quantization, LoRA, and the 8% Problem: Benchmarking Local LLMs for Production AI Iran war: We spoke to the man making Lego-style AI videos that experts say are powerful propaganda Powell, Bessent discussed Anthropic's Mythos AI cyber threat with major U.S. banks GitHub - immartian/bellamem: Persistent belief-graph memory for AI agents. Retrieves decisive context by importance — not recency, not RAG, not /compact. recursive-mode: The Repo-Native Operating System for AI Engineering After the attack on Sam Altman's home, will AI CEO's go on the offensive? The biggest advance in AI since the LLM Opus 4.6 vs GPT 5.4 One Prompt Unity World Generation Test “AI polls” are fake polls Client Challenge Can AI be a 'child of God'? Inside Anthropic's meeting with Christian leaders
GitHub - st-hannibal/arknet: Decentralized AI inference p...
st-hannibal · 2026-05-03 · via Hacker News - Newest: "AI"

arknet
Decentralized AI inference. One binary. Fair launch.

Website · Docs · Tokenomics · Explorer · Forum


Anyone with a computer earns ARK for serving AI models. Any developer queries AI through a peer-to-peer mesh — install the SDK, connect, infer.

Use the network (Python):

from arknet_sdk import Wallet, Client

wallet = Wallet.create()
wallet.save()

session = wallet.create_session(spending_limit=100_000_000, expires_secs=3600)
client = Client.connect(session=session)
response = client.infer(
    model="Qwen/Qwen3-0.6B-Q8_0",
    prompt="Hello from arknet",
    max_tokens=64,
)

Use the network (Rust):

use std::time::Duration;
use arknet_sdk::{Client, ConnectOptions, InferRequest, wallet::Wallet, session::SessionKey};

let wallet = Wallet::create();
let session = SessionKey::create(&wallet, 100_000_000, Duration::from_secs(3600))?;
let client = Client::connect(ConnectOptions {
    session: Some(session),
    ..Default::default()
}).await?;
let response = client.infer(InferRequest {
    model: "Qwen/Qwen3-0.6B-Q8_0".into(),
    prompt: "Hello from arknet".into(),
    max_tokens: 64,
    ..Default::default()
}).await?;

Your wallet address is your identity. The SDK joins the p2p mesh as a lightweight libp2p peer, discovers compute nodes via gossip, and sends inference requests directly over Noise-encrypted channels. No HTTP. No gateway. No middleman.

How it works

┌──────────┐  bootstrap  ┌───────────┐  gossip   ┌──────────┐
│  User /  │ ──────────► │ Validator │ ◄──────── │ Compute  │
│  SDK     │             │ (consensus│  "I have  │  (runs   │
│          │             │  + relay) │  model X" │  model)  │
│          │ ◄──────────────────────────────────► │          │
│          │    direct p2p (Noise-encrypted)      │          │
└──────────┘    via relay, then hole-punch        └──────────┘
                     ┌───────────┐
                     │ Verifier  │
                     │ (checks   │
                     │  output)  │
                     └───────────┘
  1. Compute nodes gossip their loaded models on the arknet/pool/offer/1 topic (heartbeat every 60s).
  2. SDK bootstraps from seed validators, subscribes to gossip, and discovers which compute nodes have which models.
  3. SDK connects to the compute node through the validator relay (libp2p circuit relay), then DCUtR hole-punches a direct connection.
  4. Inference travels directly between the SDK and compute node — the validator never sees the prompt.
  5. Verifier re-executes 5% of jobs deterministically — cheaters get slashed.
  6. Validator finalizes the receipt on L1, emission mints ARK.

Two node types: Validator (consensus + relay + seed) and Compute (inference + gossip announce). That's it.

As a developer

Install the SDK, create a wallet, create a session key, connect, infer. During the bootstrap period (first 6 months or 100 validators), inference is free.

pip install arknet-sdk    # or: cargo add arknet-sdk
from arknet_sdk import Wallet, Client

wallet = Wallet.create()
wallet.save()
session = wallet.create_session(spending_limit=100_000_000, expires_secs=3600)
client = Client.connect(session=session)

response = client.infer(
    model="Qwen/Qwen3-0.6B-Q8_0",
    prompt="Hello from arknet",
    max_tokens=64,
)
print(response.text)

Session keys authorize bounded spending from your wallet. Your main key signs a DelegationCert granting an ephemeral session key a spending_limit and expiry. Inference requests are signed by the session key — if compromised, damage is bounded.

TypeScript SDK (npm install arknet-sdk) exists but has not yet been updated for the p2p architecture. Coming soon.

As a node operator

Run the binary, pick your role, earn ARK.

curl -fsSL https://arknet.arkengel.com/install.sh | sh
arknet init
arknet start --role compute   # or: --role validator

See the full Node Operator Guide for step-by-step validator and compute setup.

Ports

Port Default Public? Purpose
P2P 26656 Yes — open in firewall Node discovery, gossip, relay, consensus, inference. Noise-encrypted.
RPC 26657 No — localhost only Operator admin. Load models, check status, submit transactions.
Metrics 9090 No — localhost only Prometheus metrics.

Discovery

The SDK discovers compute nodes through the gossip mesh:

  1. Load seed validator addresses from seeds.json (hardcoded fallback in SDK binary)
  2. Bootstrap into the libp2p mesh via a seed validator
  3. Subscribe to arknet/pool/offer/1 — receive PoolOffer messages from compute nodes
  4. Connect to a compute node offering the requested model (relay, then hole-punch)

No HTTP endpoints. No gateway. Everything is libp2p.

Wallet

Ed25519 keypair. 64-byte file at ~/.arknet/wallet.key. Address = blake3(pubkey)[0..20].

from arknet_sdk import Wallet
wallet = Wallet.create()
wallet.save()
print(wallet.address)

Or: arknet wallet create

The same wallet file works across Python, Rust, and the CLI.

CLI

# Wallet
arknet wallet create
arknet wallet balance
arknet wallet send --to 0x... --amount 1000000000

# Staking
arknet wallet stake --role compute --amount 50000000000000
arknet wallet unstake --role compute --amount 50000000000000
arknet wallet complete-unbond --role compute --unbond-id 1
arknet wallet redelegate --role compute --to-node 0x... --amount 50000000000000

# Governance
arknet governance propose --title "Add Llama 4" --body @proposal.md
arknet governance vote --proposal 1 --choice yes

# TEE
arknet tee keygen
arknet tee register --platform intel-tdx --quote-file quote.bin

Why arknet?

Centralized (OpenAI, etc.) arknet
Censorship Single kill switch Zero-trust, permissionless
Pricing Opaque, changes overnight On-chain dynamic market
Revenue Goes to one company 80% to compute, rest split across network
Models Vendor-locked Open registry, any GGUF model
Data You trust the provider Encrypted P2P, prompts never touch consensus
Confidentiality Trust the provider TEE enclaves — even host OS can't read prompts

Reward split

Every verified inference job mints ARK:

Recipient Share
Compute 80%
Verifier 7%
Treasury 5%
Burn 3%
Delegators 5%

Privacy — three tiers

Tier What it does Available
Transport Noise-encrypted P2P. Eavesdroppers see nothing. Genesis
Prompt isolation Only the assigned compute node sees your prompt. Validators and verifiers see hashes — never content. Genesis
Confidential inference (TEE) Prompts encrypted to hardware enclave (Intel TDX / AMD SEV-SNP). Even the host OS cannot read them. Genesis (protocol ready)

Fair launch

No premine. No investors. No team allocation. No airdrop. Every ARK is minted against verified inference work after genesis. The maintainer holds zero tokens at launch.

Bootstrap period (first 6 months or until 100 unique validators): inference is free, min_stake = 0, anyone can join. Compute nodes earn ARK from block emission for serving free requests.

Genesis models

10 models ship at genesis — from Raspberry Pi to server:

Tier Model Size
Edge Llama 3.2 1B 1.3 GB
Edge Qwen3.5 4B 2.7 GB
Gamer Llama 3.1 8B 4.9 GB
Gamer Qwen3.5 9B 5.7 GB
Prosumer Gemma 4 26B MoE 16.9 GB
Prosumer Qwen3.6 27B 16.8 GB
Server Qwen3.6 35B MoE 22.1 GB

Full list with SHA256 digests: models

Status

Milestone Status
Core protocol (consensus, staking, slashing) Done
Inference pipeline (escrow, compute, receipt, reward) Done
Economic layer (emission, rewards, governance, pricing) Done
Model registry + SDKs (Python, Rust) Done
P2P mesh (gossip discovery, relay, hole-punch) Done
Session keys (DelegationCert, spending limit, expiry) Done
Bootstrap emission (free-tier mints ARK) Done
Wallet CLI + block explorer Done
TEE confidential inference Done
Genesis Next

License

Apache-2.0