Cascade can now intuitively parse through and chunk up web pages and documentation, providing realtime context to the models. The key way to understand this feature is that Cascade will browse the Internet as a human would. What this means is that it will ask itself the following questions:

  1. Do I need to use the Internet to help you with your query? (ie. time-based question, live docs, user pasted URL, etc.)
  2. Which page(s) do I need to visit to get the right information and in what priority order?
  3. What parts of the page do I need to read in order to get the right information in the fastest, most efficient way possible?

Our web tools are designed in such a way that gets only the information that is necessary in order to efficiently use your credits.

This feature can potentially use credits quickly because each URL is at least 1 credit. You can toggle this feature on and off in the Windsurf Settings in the bottom right of the editor.

​
Overview

To help you better understand how Web Search works, we’ve recorded a short video describing the best practices for reducing Flow Action credits.

​
Quick Start

The fastest way to get started is to activate web search in your Windsurf Settings in the bottom right corner of the editor. You can activate it a couple of different ways:

  1. Ask a question that probably needs the Internet (ie. “What’s new in the latest version of React?”).
  2. Use @web to force a docs search.
  3. Use @docs to query over a list of docs that we are confident we can read with high quality.
  4. Paste a URL into your message.

​
Search the web

Cascade can deduce that certain prompts from the user may require a real-time web search to provide the optimal response. In these cases, Cascade will perform a web search and provide the results to the user. This can happen automatically or manually using the @web mention.

It will construct the query for you using AI and then perform a web search. This costs 1 Flow Action credit.

​
Reading Pages

Cascade can read individual pages for things like documentation, blog posts, and GitHub files. The page reads happen entirely on your device within your network so if you’re using a VPN you shouldn’t have any problems.

Pages are picked up either from web search results, inferred based on the conversation, or from URLs pasted directly into your message.

We break pages up into multiple chunks to efficiently use your Flow Action credits. It’s very similar to how a human would read a page: for a long page we skim to the section we want then read the text that’s relevant. This is how Cascade operates as well.

It’s worth noting that not all pages can be parsed. We are actively working on improving the quality of our website reading. If you have specific sites you’d like us to handle better, feel free to file a feature request!

​
Credit Usage with Web tools

Web Search uses 1 credit per search. It does not open up the search results, simply returns up to 5 links to consider reading.

Web URL reads use at least 1 credit each. If a page is long, we will break it up into chunks and present an outline to Cascade. If a section is deemed relevant, we will read that section. We try to merge chunks together as much as possible to reduce credit usage. For example, if there are 3 short neighboring chunks, we’ll merge them together so that you only use one credit to read the chunks.

For a detailed explanation of credit usage, see our video explanation.

Image UploadMemories