Skip to main content

​
Introduction

Chat has two web capabilities:
  • Web search finds current information. The model sends a query, gets back a short numbered list of results with titles, dates and snippets, and cites them as [1], [2] in its answer.
  • Web browsing opens a page the user links to, or one the model finds, and reads its text.
Both are set up under Settings → Features → Chat. You choose who runs the search and who reads the pages, and how much of each result goes to the model. Smaller results cost less input, so these limits keep web-enabled chats affordable.

​
Turning it on

  1. Pick a provider and add its API key:
    • Jina: search, page and file reading with one key.
    • Serper: Google results.
    • Perplexity: results with page excerpts included.
  2. Go to Settings → Features → Chat. In the Web search section, choose the Search provider.
  3. In the Capabilities section, turn on Web search.
  4. Open each plan that should have it and turn on Web search. To apply the change to current subscribers, save with Update snapshots. See Plans, Snapshots & Subscriptions.
Web search is unavailable until a provider is selected and its API key is set. The Web search toggle shows a warning icon while either is missing.

​
Settings

​
Reading the results

For most questions, titles and snippets are enough. When the model needs details, such as facts or figures, it asks for the text of the result pages in the same search:
  • Perplexity returns page excerpts itself.
  • Jina and Serper return snippets only. The top three result pages are then opened with the web page reader, and the search result size is shared between them.

​
Web browsing

Turn on Web browsing in the Capabilities section of Settings → Features → Chat, and in each plan that should have it.

​
Web page size

Web page size (characters) caps how much of a page the model gets when it opens a link. Longer pages are cut at this length. The default is 40,000 characters, about 10,000 tokens, and the minimum is 2,000.

​
Web page reader

The Web page reader decides who fetches and reads pages: When you choose Jina or Cloudflare, Use only as fallback appears, and it is on by default:
  • On: pages are read on your server first. Only the ones it can’t load, or reads too thin (such as JavaScript-rendered pages), go to the selected reader. This keeps costs down.
  • Off: every page goes to the selected reader first, and your server reads it only if that fails.
The same reader opens result pages for web search.
Links must point to public websites. Addresses on localhost, private networks or cloud metadata endpoints are refused, including through redirects, before any page is fetched or sent to a reader.

​
Credits

Each provider has its own rate under Settings → Finance → Credit ratios, in the Capabilities section: Pages read on your server are not charged. With Use only as fallback on, users pay for a reader only when it was needed. See Unified Credit System.

​
Troubleshooting

Check that a Search provider is selected and its API key is saved, that Web search is on under Capabilities, and that the user’s plan includes it with an up-to-date snapshot.
The capability was renamed in v5.2 and its old settings were not carried over. Turn it on again, choose a provider, and enable it in each plan with Update snapshots.
Pages built with JavaScript often have little text until a browser runs them. Choose Jina or Cloudflare as the web page reader, with Use only as fallback on.

​
Need Help?

If you need assistance with Aikeedo:

Professional Support

Get expert help from our team with a paid support subscription

Troubleshooting Guide

Check common issues and solutions