🌐 agentkit-cleanweb

Official Coinbase AgentKit Action Provider for CleanWeb Studio
Pure B2A (Business-to-Agent) 87% Token Noise Reduction, Research PDF Parser, YouTube Gemini Flash Intelligence, and Base Mainnet EIP-712 Ground Truth Attestations.

🔗 Official CleanWeb Multi-Registry Suite (npm ↔ PyPI Companion Packages)

For full-stack multi-agent deployments, pair agentkit-cleanweb with the official Python Core / MCP Server (PyPI) and ElizaOS v2 Plugin (npm):

| Registry | Package | Role in Your Agent Stack | Install Command | | :---: | :--- | :--- | :--- | | npm | agentkit-cleanweb | Coinbase AgentKit Action Provider (Base Mainnet EIP-712 & x402 Vaults) | npm install agentkit-cleanweb @coinbase/agentkit | | PyPI | x402-cleanweb-agent | Python Core Engine, Claude/Cursor MCP Server & Dual-Core Eliza Brain | pip install x402-cleanweb-agent (or uvx x402-cleanweb-agent) | | npm | cleanweb-eliza-plugin | ElizaOS v2 Plugin (Solana x402 Micropayments & A.GRID Priority 0 Dispatch) | npm install cleanweb-eliza-plugin |

📌 Why CleanWeb for Coinbase AgentKit?

When autonomous AI agents crawl or scrape the live web, raw HTML boilerplate (cookie banners, JavaScript scripts, navigation clutter, ads) consumes 70% to 90% of the LLM context window.

On GPT-4o or Claude 3.5 Sonnet, scraping a single 80KB page costs ~$0.05 in input tokens.

CleanWeb reduces web content noise by 87%, transforming raw HTML into clean, LLM-ready structured Markdown and delivering 86.2% net cost savings on every downstream agent execution.

✨ Key Features

• 87% Token Compression: 80KB HTML (20,000 tokens) ➔ 2,600 tokens clean Markdown.
• On-Chain EIP-712 Proofs: Cryptographically signed attestation hashes on Base Mainnet (Chain ID: 8453).
• Research PDF Parsing: Whitepapers, financial 10-K filings, and arXiv research parsed in < 2 seconds.
• YouTube Video Intelligence: Instant subtitles and Google Gemini 3.8 Flash bullet-point summaries.
• Gasless x402 Micropayments: Pre-funded USDC session vaults on Base with < 1ms execution latency.

🚀 Quickstart

1. Installation

# TypeScript / Coinbase AgentKit
npm install agentkit-cleanweb @coinbase/agentkit

# Companion Python Core & MCP Server (Optional: for local self-hosting & Claude/Cursor MCP)
pip install x402-cleanweb-agent

2. Register Action Provider with AgentKit

import { AgentKit, EvmWalletProvider } from "@coinbase/agentkit";
import { cleanWebActionProvider } from "agentkit-cleanweb";

// 1. Initialize your Coinbase AgentKit EVM Wallet Provider
const walletProvider = new EvmWalletProvider({
  privateKey: process.env.CDP_WALLET_PRIVATE_KEY,
  networkId: "base-mainnet", // or "base-sepolia"
});

// 2. Initialize CleanWeb Action Provider
const cleanWeb = cleanWebActionProvider({
  vaultKey: process.env.AGENT_VAULT_KEY, // Optional: Pre-funded USDC vault key
});

// 3. Register with AgentKit
const agentKit = await AgentKit.from({
  walletProvider,
  actionProviders: [
    cleanWeb,
    // ...other action providers (e.g. security-gate, weth, erc20)
  ],
});

// 4. Retrieve actions and bind to your LangChain / ElizaOS agent
const actions = agentKit.getActions();
console.log(`Registered ${actions.length} CleanWeb actions!`);

🛠️ Provided Actions

| Action Name | Description | Base Pricing | | :--- | :--- | :---: | | cleanweb_clean_web_content | Scrapes any webpage URL and compresses noise/ads into 87% token-reduced Markdown. | 0.001 USDC | | cleanweb_clean_pdf_research | Parses online whitepapers, arXiv research, or financial PDFs into structured text. | 0.005 USDC | | cleanweb_clean_youtube_transcript | Extracts subtitles and Gemini 3.8 Flash summary from YouTube videos without downloading. | 0.005 USDC | | cleanweb_oracle_grounding | Factual discovery with on-chain verifiable EIP-712 cryptographic signature for Base (8453). | 0.035 USDC | | cleanweb_manage_x402_vault | Checks USDC balance, retrieves Base deposit address, or checks agent VIP pass status. | Free |

💡 Example: Standalone Usage (Without Full AgentKit)

All actions can also be called standalone in any Node.js / TypeScript environment:

import { cleanWebContent, oracleGrounding } from "agentkit-cleanweb";

// 1. Clean a Webpage with 87% token compression
const webResult = await cleanWebContent({
  url: "https://news.ycombinator.com",
});
console.log(`Title: ${webResult.title}`);
console.log(`Token Reduction: ${webResult.tokenReductionRate}`);
console.log(`Clean Markdown:\n${webResult.cleanMarkdown.slice(0, 300)}...`);

// 2. Generate Base EIP-712 Attestation Proof
const proof = await oracleGrounding({
  query: "Base network daily transactions ATH",
  chainId: 8453,
});
console.log(`Verdict: ${proof.verdict}`);
console.log(`EIP-712 Signature: ${proof.attestation?.signature}`);

🔒 Dual-Core Security Gate & ElizaOS Brain Integration

CleanWeb is powered by the Dual-Core Cognitive Fortress (x402-cleanweb-agent on PyPI), ensuring that all scraped content passes through a 5-stage zero-trust pipeline:

1. Indirect Prompt Injection & Steganography Neutralization
2. AST Malicious Code Execution Firewall
3. ElizaOS Brain Causality & Brier Score Ground-Truth Verification
4. Double Cryptographic Seal (Keccak-256 Merkle Root + EIP-712 Signature)

📄 License

MIT © CleanWeb & A.GRID Protocol