How to use ScrapeGraphAI with Ollama for free

What you need

• Windows 10 or 11
• Python 3.12 or newer—the current ScrapeGraphAI package requires at least 3.12
• Ollama and enough free storage for the model and browser download
• A public webpage you are permitted to access
• Comfort copying commands and creating one Python file

Set it up

1. Install Ollama and confirm it opens

   Download the normal Windows installer from Ollama's official site. Open a new PowerShell window after installation and confirm the command is available.

   ollama --version

2. Download the local AI model

   ScrapeGraphAI's current official example uses Llama 3.2. Pull it once, then send a tiny test prompt. The model remains on the computer for later runs.

   ollama pull llama3.2
   ollama run llama3.2 "Reply with the single word ready"

3. Create an isolated Python project

   Install Python 3.12 from Python's official Windows installer if needed. These commands create a separate environment and call its Python directly, avoiding PowerShell activation-policy problems.

   mkdir scrapegraph-local
   cd scrapegraph-local
   py -3.12 -m venv .venv
   .\.venv\Scripts\python.exe -m pip install --upgrade pip
   .\.venv\Scripts\python.exe -m pip install scrapegraphai
   .\.venv\Scripts\python.exe -m playwright install

4. Create the first scraper

   Create a file named scrape.py inside scrapegraph-local and paste this example. It uses Books to Scrape, a public practice site made for scraping exercises, and disables ScrapeGraphAI's documented anonymous telemetry before importing the package.

   import os
   os.environ["SCRAPEGRAPHAI_TELEMETRY_ENABLED"] = "false"
   
   import json
   from pathlib import Path
   from scrapegraphai.graphs import SmartScraperGraph
   
   config = {
       "llm": {
           "model": "ollama/llama3.2",
           "model_tokens": 8192,
           "format": "json",
       },
       "verbose": True,
       "headless": True,
   }
   
   scraper = SmartScraperGraph(
       prompt=(
           "Extract the first 10 visible books. For each book return "
           "only its title and displayed price. Do not guess missing values."
       ),
       source="https://books.toscrape.com/",
       config=config,
   )
   
   result = scraper.run()
   output = json.dumps(result, indent=2, ensure_ascii=False)
   print(output)
   Path("result.json").write_text(output, encoding="utf-8")

5. Run it and open the result

   Keep Ollama running, then execute the script. The first run can take longer while the browser and model warm up. Open result.json when the command finishes.

   .\.venv\Scripts\python.exe scrape.py

6. Change one thing at a time

   Replace the prompt first and rerun the safe practice page. Once the fields are reliable, replace the source with a public page whose rules permit automated access. Keep the request small while measuring accuracy and speed.

Verify it works

Do not call the setup finished until these small tests pass:

• ollama list shows llama3.2
• The script finishes without asking for an API key or ScrapeGraphAI credits
• result.json contains titles and prices visible on the practice page
• A second run returns the same essential facts
• Every extracted value has been checked against the source

Protect the service

• Use only pages you are authorized to access and respect terms, robots instructions and rate limits
• Never use the setup to bypass logins, paywalls, CAPTCHAs or other access controls
• Local AI does not hide requests from the target website
• Do not collect sensitive personal information merely because it is technically visible
• Treat AI output as an unverified draft, especially for prices, dates and names

Official resources

Videos are supplemental. Use the official documentation for current commands, versions and security guidance.