BotBrowser

Token-efficient web browser for LLM agents

---

A typical web page is 50,000+ tokens. The useful content? 2,000–5,000 tokens.

BotBrowser strips the bloat and gives your agents clean markdown — saving 90–95% of tokens.

Raw HTML:   52,000 tokens  ████████████████████████████████████████████████████
BotBrowser:  3,200 tokens  ██████
                           ↑ 94% savings

Install

npm install botbrowser    # JavaScript / TypeScript

pip install botbrowser    # Python

No API key. No server. No config. Just install and extract.

Quick Start

// JavaScript / TypeScript
import { extract } from 'botbrowser';

const result = await extract('https://example.com/article');
console.log(result.content);       // clean markdown
console.log(result.metadata.tokenSavingsPercent);  // 94

# Python
from botbrowser import extract

result = extract("https://example.com/article")
print(result.content)                         # clean markdown
print(result.metadata.token_savings_percent)  # 94

What You Get Back

{
  "url": "https://example.com/article",
  "title": "Article Title",
  "description": "Meta description",
  "content": "# Article Title\n\nClean markdown content...",
  "textContent": "Plain text version...",
  "links": [
    { "text": "Related Article", "href": "https://example.com/related" }
  ],
  "metadata": {
    "rawTokenEstimate": 52000,
    "cleanTokenEstimate": 3200,
    "tokenSavingsPercent": 94,
    "wordCount": 1250,
    "fetchedAt": "2026-02-26T10:30:00.000Z"
  }
}

Why BotBrowser?

• Token-first — Built specifically to minimize LLM token usage. Every design decision optimizes for fewer tokens while preserving meaning.
• Dual native SDKs — Real implementations in both JS and Python, not thin wrappers. Use whichever fits your stack.
• Zero setup — npm install or pip install. No API key, no account, no server to run. Works offline.
• Battle-tested extraction — Mozilla Readability (JS) and Trafilatura (Python) — the same engines powering Firefox Reader View and academic web research.
• Open source — MIT licensed. Self-host, fork, embed, do what you want.

How It Works

URL → Fetch → Extract → Clean → Markdown

1. Fetch — Smart HTTP with user-agent rotation, redirect handling, timeouts
2. Extract — Identifies main content using Readability (JS) / Trafilatura (Python)
3. Clean — Strips scripts, styles, ads, nav, footers, cookie banners, tracking, hidden elements
4. Convert — Clean Markdown preserving headings, lists, links, tables, code blocks

Options

const result = await extract({
  url: 'https://example.com',
  format: 'text',          // "markdown" (default) or "text"
  timeout: 10000,          // request timeout in ms (default: 15000)
  includeLinks: false,     // extract links (default: true)
});

result = extract(
    "https://example.com",
    format="text",
    timeout=10000,
    include_links=False,
)

REST API (Optional)

For language-agnostic access or shared infrastructure:

docker compose up
# or: cd js && pnpm install && pnpm build && pnpm dev

curl -X POST http://localhost:3000/extract \
  -H 'Content-Type: application/json' \
  -d '{"url": "https://example.com"}'

Python client for the REST API:

from botbrowser import BotBrowserClient

client = BotBrowserClient("http://localhost:3000")
result = client.extract("https://example.com")

Development

# JS
cd js && pnpm install && pnpm build && pnpm test

# Python
cd python && pip install -e ".[dev]" && pytest tests/ -v

License

MIT