Add Python actor templates for LLM-based web scraping frameworks

[vdusek]

Summary

Add new Python actor templates for modern LLM-based web scraping frameworks. These frameworks leverage large language models for intelligent data extraction, making it easier to scrape unstructured or semi-structured web content.

Templates to add

• Crawl4AI – LLM-friendly web crawler and scraper with built-in support for structured extraction
• Scrapling – high-performance, adaptive web scraping library with intelligent content extraction
• Browser Use – AI agent framework for browser automation using LLMs
• ScrapeGraphAI – LLM-powered scraping pipelines using graph-based logic

Notes

• Each template should follow the existing Python template conventions in this repo
• Templates should include a reasonable default use case / example configuration
• Consider which base Docker image is most appropriate for each (e.g. Playwright-based for browser-heavy frameworks)

🤖 Generated with Claude Code

[m13v]

For the Browser Use template specifically, here are some practical defaults that would save users a lot of trial and error:

Set the browser context to reuse cookies between navigation steps. Browser Use agents frequently need to log in, then navigate to authenticated pages. If each step starts with a fresh context, the agent wastes tokens re-authenticating.

For the Docker image, use the Playwright base with Chromium only (skip Firefox and WebKit). The full Playwright image with all browsers is 2GB+; Chromium-only cuts it to ~800MB, and 99% of scraping tasks only need Chromium.

For the Crawl4AI template, the default extraction mode should be "markdown" rather than "raw HTML." Markdown is much more token-efficient when you're feeding the extracted content back into an LLM for analysis. This alone can cut API costs by 3-4x on content-heavy pages.

One gotcha with ScrapeGraphAI on Apify: their graph nodes make synchronous HTTP calls by default, which blocks the event loop. The template should wrap extraction nodes in asyncio.to_thread() or use the async variants to play nicely with Apify's async runtime.

For all templates, I'd suggest including a built-in rate limiter (e.g., max 2 requests/second by default). LLM-guided scraping tends to be slower per-page than traditional scraping, so users often crank up concurrency to compensate, which gets them IP-banned fast.

[m13v]

For reference on browser session management with concurrent agents: https://github.com/m13v/browser-lock/blob/main/playwright-lock.sh

And here's how our AI agent handles tool execution including browser automation calls: https://github.com/m13v/fazm/blob/main/Desktop/Sources/Providers/ChatToolExecutor.swift