---
title: "Web Search"
sidebar_position: 0
slug: /features/chat-conversations/web-search
---
import ThemedImage from '@theme/ThemedImage';
import useBaseUrl from '@docusaurus/useBaseUrl';
import DocCardList from '@theme/DocCardList';
# Web Search
Web search lets the AI look up information online to help answer your questions.
There are two jobs involved: **finding pages** and **reading pages**. The **search engine finds pages**. The **web loader reads them**.
## Search engine vs. web loader
Think of looking up a recipe online. You type “chocolate cake recipe” into Google, then click a result to read the recipe. Those are two separate steps. Open WebUI does something similar:
| Setting | What it does | What the AI gets |
| --- | --- | --- |
| **Web Search Engine** | Searches for “chocolate cake recipe.” | A list of links to recipe pages, with short previews. |
| **Web Loader Engine** | Opens one of those links and gets the page's text. | The actual recipe: ingredients and instructions. |
**“Web loader” just means “web page reader.”** A search result usually contains a link and a short preview. The loader gets the text from the page itself when the AI needs more detail. It can also read a link you paste into the chat.
That is why there are two settings: one chooses **who finds the links**, and the other chooses **how to read the pages behind those links**.
The web loader settings apply with web search off too. **Web Search Engine** is required only while **Web Search** is on, so the **Loader** settings save without an engine. While **Web Search** is off, the page hides **Web Search Confirmation**, **Web Search Engine** and its settings, **Search Limits**, **Fetch URL Content Length Limit**, **Domain Filter List** and **Bypass Web Loader**. Their saved values are kept. **Bypass Embedding and Retrieval** and **Trust Proxy Environment** stay visible.
### Playwright and External loader settings
Choosing `playwright` or `external` as the **Web Loader Engine** shows extra fields in the **Loader** section.
- **Playwright WebSocket URL** (`PLAYWRIGHT_WS_URL`): Open WebUI connects to the Chromium browser at this address instead of starting its own. The Docker image does not include Chromium. The container installs it at startup only when the environment sets `WEB_LOADER_ENGINE=playwright` and leaves `PLAYWRIGHT_WS_URL` empty. Choosing `playwright` in the admin panel alone installs no browser, so set the environment variable and restart, or enter a WebSocket URL.
- **Playwright Timeout (ms)** (`PLAYWRIGHT_TIMEOUT`, default `10000`): the time in milliseconds Open WebUI gives each page to load, and each request the page makes while loading. A page that runs out of time is skipped and logged as `Error loading :` followed by Playwright's error. The other pages still load. If no page loads, web search fails with `[ERROR: None of the web search results could be loaded]` (unless **Bypass Embedding and Retrieval** is on), and a link attached with `#` fails with `[ERROR: Could not read content from ]`.
- **External Web Loader URL** (`EXTERNAL_WEB_LOADER_URL`): Open WebUI sends a `POST` to this URL with the JSON body `{"urls": [...]}`, up to 20 URLs per request, and the header `Authorization: Bearer `. The service must answer with a JSON array of objects, each with `page_content` (the page text) and `metadata`. Set `metadata.source` to the URL the text came from: web search keeps only the results whose link matches a returned `source`. An error status or a response that is not such an array skips the whole batch, logged as `Error extracting content from batch : `. Open WebUI sets no timeout on this request, so a service that never answers holds up the search.
## Why are there so many options?
Open WebUI lets you choose which service does each job. The lists are **alternatives to choose from**. You do **not** need to set up all of them.
- **Search engines offer different ways to find results.** Some work without an account, some require an account and may charge for searches, and some can run on your own server. They can also return different results for the same question.
- **Web loaders offer different ways to read pages.** Websites are built differently: some provide text straight away, while others need a browser to run the page before the text appears. Different loaders handle these situations differently.
Some services, such as Tavily and [Exa](/features/chat-conversations/web-search/providers/exa#4-optional-use-exa-to-read-pages), can do both jobs, so you may see the same name in both lists. You can also use different services for the two jobs.
## What should I choose to get started?
:::tip Search engine quality
For the strongest search results, start with a hosted search API built for LLM workflows, such as [Exa](/features/chat-conversations/web-search/providers/exa), [Tavily](/features/chat-conversations/web-search/providers/tavily), [SearchApi](/features/chat-conversations/web-search/providers/searchapi), [Serper](/features/chat-conversations/web-search/providers/serper), [SerpApi](/features/chat-conversations/web-search/providers/serpapi), [Brave](/features/chat-conversations/web-search/providers/brave), or [Kagi](/features/chat-conversations/web-search/providers/kagi). [SearXNG](/features/chat-conversations/web-search/providers/searxng) is still useful when you want a self-hosted, privacy-preserving metasearch engine, but result quality depends heavily on its upstream engines, rate limits, and instance configuration.
:::
1. **Choose one search engine and follow its setup guide below.** A hosted search API is a service Open WebUI connects to so it can search on your behalf. These services usually require an account and an **API key**: a code that connects Open WebUI to your account with the service.
2. **Leave the web loader at its default setting.** You do not need a separate page-reading service to start. If pages cannot be read, the [troubleshooting guide](/troubleshooting/web-search) can help you decide whether to change the loader. The [loader settings reference](/reference/env-configuration#web-loader-configuration) explains the available options.
:::tip Search works, but the page content is empty?
Finding a page does not guarantee Open WebUI can read it. A website may appear in search results but block tools from opening it, or load its text in a way your loader cannot read.
:::

:::tip Require confirmation before searching
Admins can require users to **confirm before a web search runs**, so a query is never sent to an external provider unintentionally. Enable it in **Settings → Web Search** (admin section), where the switch shows while **Web Search** is on (or via [`ENABLE_WEB_SEARCH_CONFIRMATION`](/reference/env-configuration#enable_web_search_confirmation)); the prompt text is set by [`WEB_SEARCH_CONFIRMATION_CONTENT`](/reference/env-configuration#web_search_confirmation_content).
The text goes in **Web Search Confirmation Content**, which shows below the switch while it is on. Users see it in a `Use Web Search?` dialog with **Continue** and **Cancel** when they turn on **Web Search** in the chat input, and when they send a message while web search is on and not yet confirmed. With the field empty, the dialog shows `Your query will be sent to the configured web search provider.` One confirmation covers the rest of the chat. Starting a new chat or turning web search off asks again. The check runs in the chat interface only, so requests sent straight to the API are not held for confirmation.
:::
:::info Web search runs with the chat toggle off
**Web Search in Chat** in **Settings → Interface** switches between **Default** and **Always**. With **Always**, every message uses web search, even after you turn off **Web Search** in the chat input. The only exception is a selected model whose **Web Search** capability is off. To turn web search on and off per chat, set **Web Search in Chat** back to **Default**.
:::