​
/AskLLM

Essentially sends a prompt to an LLM like GPT or Claude and gives you back the answer. Useful for generating content, summarizing, categorizing, answering questions, and anything else you’ve done with chatbots before. You can specify the model you want to use. Popular options include Sonnet 3.5 (default), GPT-4o, and Perplexity. You can be more specific by specifying the model ID, such as claude-3-5-sonnet. Here’s the full list of model IDs for OpenAI, Anthropic, and Perplexity.

​
Tips

  • It’s often helpful to give examples of the kind of output you’re looking for.

​
Example usage

/AskLLM to read Article Text and tell me the 3 main points. Output just the 3 points, nothing else.

​
/Google

Queries both Google and Bing, returning up to 4 results from each. Note that this only grabs link results, not the other widgets that show up on the search results page. Each link comes with the page title and a snippet of the page. Supports advanced search operators like site: and inurl:.

​
Tips

  • Combine with /Visit and /Extract for comprehensive web research.

​
Example usage

/Google “site:techcrunch.com AI startups funding 2023” and output the first result.

​
/Visit

Fetches webpage content from specified URLs or search results. Can process multiple URLs in parallel.

​
Tips

  • The input to this is usually either a URL column or /Google search results. /Extract is almost always used afterwards.
  • If you’re familiar with HTML and CSS selectors, you can ask the agent to grab data from the webpage using those selectors.

​
Example usage

/Visit Blog post URL and /Extract the paragraph mentioning Matrices.
If you’re technical, you can try something like this:
/Visit Product URL and output the text content of the HTML element with class name “price”.

​
/FilterVisitedWebpages

This is useful when you want to narrow down search results before extracting info from them. It ranks webpages by how well they match the criteria you provide, and then eliminates webpages that don’t meet the criteria at all.

​
Tips

  • Must be used after /Visit.
  • Only use this when the only way to filter is to have AI look at the actual content of the webpage. Don’t use this when you can filter by URL or simple keyword matching.
  • This is particularly useful when you want to output the URL of the single most relevant webpage.

​
Example usage

/Visit all the URLs in URLs list and /FilterVisitedWebpages to ones that contain Korean text.
  1. /Google “Name Company School City”
  2. /Visit all results
  3. /FilterVisitedWebpages to ones that seem to detailed profiles about the person Name who works at Company, went to School, and currently lives in City. Output the URL of the single most relevant result.

​
/Extract

Your go-to tool for extracting or analyzing information in any webpage or file. Note: It currently only sees text, not images or other media. When given a file, it performs OCR on it.

​
Tips

  • You must /Visit a webpage before you can extract from it.
  • This does not need to output the content exactly as it appears in the source. You can ask it to reformat the content however you want. It helps to provide example outputs!
  • If you want structured (JSON, if you’re technical) output, you can change the data type of your column to “JSON” in the column settings, and then prompt for the exact format you want.

​
Example usage

/Visit Job Posting URL and /Extract the following information:
  1. Job title
  2. Required years of experience
  3. List of required skills
  4. Salary range (if available)
  5. Application deadline

​
/FetchPersonData

Sometimes, information about individual people, such as their email addresses, can be hard to find on the web. Use this tool to grab info like that from data enrichment providers. Can fetch emails, phone numbers, work history, educational history, and social URLs.

​
Tips

  • Names are rarely unique. To precisely identify a person, please ideally provide a LinkedIn URL, or some combo of information like their name + the company they work at.

​
Example usage

Use /FetchPersonData to grab the email and phone number of the person with LinkedIn URL LinkedIn Profile URL

​
/FindPersonInOrganization

Find a person at a specified company with a given job title.

​
Tips

  • Sometimes multiple companies have the same name, so if you can, provide notes on how to disambiguate (such as “fintech companies only”).
  • Job titles are rarely exact, so you can use fuzzy guidelines like “find the most senior operations person”.

​
Example usage

Use /FindPersonInOrganization to find the head of security at crypto company Company name.