AI Mode

AI Mode turns a web page and a simple description of the data you need into a reusable data extraction robot. Just enter a URL and a prompt, such as “Extract the five men's basketball shoes with their names, prices, ratings, and product URLs.” Maxun figures out how to extract the requested data and builds a robot that can run repeatedly.

LLM calls are only needed when creating the robot, not on every run. Once created, the robot can extract data from the page efficiently without recurring AI costs. You can schedule runs, monitor changes, and use your robots through the API, SDK, and CLI.

AI Mode is part of the self-hosted open-source edition and uses the LLM you configure. On Maxun Cloud, robots built from a prompt are Agents. See Agent vs AI Mode for how they differ.

Create a robot with AI Mode

1. Open Robots → Create Robot → Extract and switch to AI Mode.
2. Website URL (required): the page that holds the data. AI Mode reads this page only.
3. Prompt: describe the data, the fields you want and, if it matters, how many items. For example:
   • "Extract the first 20 product names and prices"
   • "Get company names, descriptions and batch"
   • "Extract article headlines, authors and publication dates"
4. Model settings: choose the LLM provider and model. See LLM providers.
5. Click Create and run. Maxun builds the robot, runs it once and opens the results.

How it works

AI Mode turns a URL and a natural-language description of the data you need into a reusable data extraction robot.

1. Specify the page and the data. Enter the URL of the page containing your data and describe what you want to extract, including specific fields or a desired number of items.

2. Understand your request. Maxun interprets your instructions to identify the information you need, rather than simply extracting the largest list or collection on the page.

3. Extract the data. Maxun identifies the requested information on the specified page and builds a robot to extract it. It can reveal content hidden behind cookie banners, modals, accordions, or lazy loading. If you request more items than are initially visible, it can use the page's own pagination or load-more controls to retrieve them.

4. Validate and save the robot. Maxun checks the extraction and saves a reusable robot that can run against your specified URL. If some requested fields aren't available, the robot records what's missing. If the requested data cannot be found, creation stops with an explanation.

5. Run it whenever you need. Once created, the robot can be run repeatedly without making LLM calls on every run. Schedule it, monitor extraction results, or integrate it through the API, SDK, and CLI.

Designed for reliable extraction, not autonomous browsing

AI Mode is built to create focused data extraction robots, not autonomous agents that explore the web to complete open-ended tasks.

• Page-specific: Extracts data from the URL you provide, rather than searching other websites for missing information.
• Predictable: Stays within the specified page and its permitted interactions. It doesn't independently browse elsewhere or follow unrelated links.
• Intent-driven: Extracts the fields and items you describe, rather than guessing what data might be useful.
• Reusable and cost-efficient: AI helps build the robot; subsequent runs reuse it without recurring LLM calls for extraction.

If the information isn't available on the specified page, Maxun tells you instead of silently searching elsewhere or inventing missing values.

LLM providers

Choose a provider under Model settings when you create the robot, or pass it from the SDK, API or CLI.

Provider	API key	Default model	Notes
Anthropic	Required: in Model settings or ANTHROPIC_API_KEY	claude-sonnet-5	Get a key from the Anthropic Console
OpenAI / OpenAI-compatible	Required: in Model settings or OPENAI_API_KEY	gpt-5.6-terra	Set a base URL to use any OpenAI-compatible server
Ollama	Not needed locally. Ollama Cloud: OLLAMA_API_KEY	llama3.2	Pull the model first (ollama pull llama3.2), or set another one

Choosing a model in Model settings overrides the default. Any model that follows instructions and returns JSON reliably works; larger models give better results on complex pages.

Ollama

• Local: Maxun connects to http://localhost:11434. If Maxun runs in Docker and Ollama runs on your machine, set OLLAMA_BASE_URL=http://host.docker.internal:11434.
• Ollama Cloud: set the base URL, your API key, and a model that Ollama Cloud serves:

OLLAMA_BASE_URL=https://ollama.com
OLLAMA_API_KEY=your-ollama-api-key
OLLAMA_MODEL=gpt-oss:20b

If the model doesn't exist on the server, robot creation fails with Ollama's own message (for example model 'llama3.2' not found) and what to check.

See Environment Variables for all settings.

When to use AI Mode

• You want data from a page without learning the Recorder.
• The data is a list (products, jobs, articles, listings) or a few values on a page (a price, an address, a date).
• You know the page that shows the data.

When to use Recorder Mode instead

• You need to log in, fill in forms or click through several steps before the data appears.
• The data is spread across several pages, or needs the site's search, filters or sorting applied first.
• You need exact control over every step of the robot.

Using with the SDK

const robot = await maxun.extract('YC companies', 'https://www.ycombinator.com/companies', {
  prompt: 'Extract the first 20 company names and descriptions',
  llmProvider: 'anthropic',
  llmApiKey: 'your-llm-api-key',
});
const result = await robot.run();

The URL is required. See Node.js Extract and Python Extract.

Using with the CLI

maxun robots extract \
  -p "Extract all product names and prices" \
  -u "https://example.com/shop" \
  -n "Shop Extractor"

See CLI robots.

Writing effective prompts

Name the fields you want and, if it matters, how many items. For more examples, see LLM Extraction Prompts.