🧹

CLEANING

cleaning.4bots.ai

Any URL → clean markdown. Strips nav, cookie banners, ads, and JS cruft. A 50k-token HTML page becomes 2–5k tokens. Your agent spends less, learns more.

Live Free — 25/day 99% Token Savings x402 Native MCP Compatible

URL in. Clean markdown out. Agents save 99% on tokens.

Every token your agent wastes parsing a cookie banner is a token you pay for. CLEANING fetches any URL, strips the junk, and returns clean markdown — so your agent gets the content, not the clutter.

🧹 Boilerplate removal

Strips navigation bars, footers, sidebars, cookie consent banners, ads, tracking pixels, and social widgets. What's left is the actual content.

📊 Token estimation

Every response includes a token estimate. Your agent can decide whether to fetch the full page or just the summary — before spending a cent.

⚡ Smart caching

Same URL requested 100 times = 1 fetch. 1-hour TTL per URL with LRU eviction. Cached responses return in under 50ms.

🤖 Agent-native output

Clean markdown, structured JSON, MCP-compatible. No HTML parsing. No guesswork. Ready for your context window.

Open-source extraction. Battle-tested reliability.

CLEANING uses trafilatura — the state-of-the-art open-source content extraction library. No proprietary black boxes. No API keys. Just clean, structured output from any URL.

Primary

trafilatura

Python library for web content extraction. Handles 95%+ of web pages correctly. Extracts main content, strips boilerplate, preserves structure (headings, lists, links). Apache 2.0 license. The gold standard for open-source extraction.

HTTP

httpx

Modern async HTTP client. Follows redirects (up to 5 hops), respects robots.txt, handles gzip/brotli decoding. Fetches the raw HTML that trafilatura processes.

Caching

SQLite

Local cache with 1-hour TTL per URL. LRU eviction when the cache fills. Same URL from any agent = one fetch, many serves. Reduces load on target sites and speeds up responses.

LLM (Paid Tier)

GX10 — Qwen3.6-27B-FP8

Local inference server for LLM-enhanced extraction. Summarization, field extraction, and token-budget compression. Zero marginal cost — runs on our own hardware. Paid tier only.

One endpoint. Any URL. Clean output.

GET /v1/free/clean?url=:url

Fetch any URL and return clean markdown. Free tier: 25/day. Paid: up to 500/day.

curl "https://cleaning.4bots.ai/v1/free/clean?url=https://example.com"

Example response:

{
  "url": "https://example.com",
  "title": "Example Domain",
  "cleaned_markdown": "# Example Domain\n\nThis domain is for use in illustrative examples...",
  "token_estimate": 847,
  "model": "trafilatura",
  "cached": false,
  "free_remaining": 24
}

GET /v1/free/clean?url=:url (with payment)

Include a valid payment header for rule-based or LLM-enhanced tier. Same endpoint, upgraded output.

curl -H "PAYMENT-SIGNATURE: <base64-payload>" \
  "https://cleaning.4bots.ai/v1/free/clean?url=https://example.com"

GET /health

Health check. Always free, always returns 200.

curl "https://cleaning.4bots.ai/health"

Free to try. Pennies to scale.

25 free requests per day covers development and light agent workloads. Paid tiers unlock higher limits, JS-rendered pages, and LLM-enhanced extraction.

Free
$0 / month
No wallet required
  • 25 requests / day
  • Clean markdown output
  • Token estimate in response
  • Title extraction
  • 1-hour cache
  • robots.txt respect
Rule-Based
$0.001 / request
via x402 USDC on Base
  • 500 requests / day
  • Everything in Free
  • Higher rate limits
  • Priority caching
  • Faster timeouts (15s)
LLM-Enhanced
$0.003 / request
via x402 USDC on Base
  • 200 requests / day
  • Everything in Rule-Based
  • Summarization on demand
  • Structured field extraction
  • Token budget compression
  • JS-rendered pages (Playwright)

Clean a URL now.

No signup. No API key. Just send a request.

How x402 works