Crawl4AI Proxy for Open WebUI

A lightweight proxy that bridges Open WebUI's external web loader API to Crawl4AI, enabling high-quality page content fetching for AI chat sessions.

Why?

Open WebUI's default web loader struggles with JavaScript-heavy pages and bot detection. This proxy uses Crawl4AI's headless browser to fetch clean markdown content from any URL, dramatically improving the quality of web-search-enhanced AI responses.

Architecture

Open WebUI → Crawl4AI Proxy (content fetch) → Crawl4AI (markdown extraction)

1. Open WebUI sends URLs to this proxy as the external web loader
2. The proxy calls Crawl4AI /md endpoint with filter=fit to extract clean content (strips navigation, ads, sign-in prompts). If the filtered result is too short (<100 chars), it retries without the filter as a fallback
3. Content is returned to Open WebUI and injected into the AI model's context

Prerequisites

• Crawl4AI running (Docker or standalone) — unclecode/crawl4ai
• Open WebUI — open-webui/open-webui

Quick Deploy (Docker)

docker run -d --name crawl4ai-proxy -p 8087:8087 -e CRAWL4AI_URL=http://<your-crawl4ai-ip>:11235/md zkkzkk32312/crawl4ai-proxy

Environment Variables

Unraid Deployment

Via Community Apps (recommended)

1. Go to Apps tab in Unraid
2. Search for Crawl4AI Proxy
3. Click Install
4. Configure the settings:
   • Proxy Port — host port (default 8087)
   • Crawl4AI URL — your Crawl4AI /md endpoint
   • Proxy API Key — optional, leave empty to disable
5. Click Apply

Manual

1. Copy templates/crawl4ai-proxy.xml to /boot/config/plugins/dockerMan/templates-user/
2. Go to Docker → Add Container in Unraid
3. Select my-crawl4ai-proxy template
4. Configure the settings and click Apply

Open WebUI Configuration

1. Go to Admin Panel → Web Loader
2. Set Web Loader Engine to external
3. Set External Web Loader URL to: http://<proxy-ip>:8087/search
4. If you set PROXY_API_KEY, enter it in External Web Loader API Key

Note: Make sure Bypass Web Loader is OFF in Admin Panel.

API

POST /search

Fetches page content for a batch of URLs in parallel (all URLs are fetched simultaneously via asyncio.gather).

Request:

{
  "urls": ["https://example.com", "https://github.com/..."]
}

Headers:

Authorization: Bearer <api_key>   (if PROXY_API_KEY is set)

Response:

[
  {
    "page_content": "# Page Title\n\nExtracted markdown content...",
    "metadata": { "source": "https://example.com" }
  }
]

GET /health

Health check endpoint.

{ "status": "ok" }

License

MIT