Skip to content

Web Search

Let the model search the internet and fetch web pages in real time. When enabled, the model can look up current information, verify facts, retrieve documentation, or research topics — all autonomously via tool calling.

Supported Providers

Provider Description Free Tier Setup
Brave Privacy-focused search API Yes (2,000 queries/month) api.search.brave.com
Serper Fast Google Search API Yes (2,500 queries/month) serper.dev
Tavily AI-optimized search, built for LLM agents Yes (1,000 queries/month) tavily.com
SearXNG Self-hosted metasearch engine Self-hosted (unlimited) Your own instance

Setup

Brave

  1. Get an API key from Brave Search API
  2. In Agora, go to Settings → Web Search
  3. Select Brave as the search provider
  4. Paste your API key

Serper

  1. Get an API key from serper.dev
  2. In Agora, go to Settings → Web Search
  3. Select Serper
  4. Paste your API key

Tavily

  1. Get an API key from tavily.com
  2. In Agora, go to Settings → Web Search
  3. Select Tavily
  4. Paste your API key

SearXNG

  1. Set up a SearXNG instance (self-hosted) or use a public instance
  2. In Agora, go to Settings → Web Search
  3. Select SearXNG
  4. Enter your instance's Base URL (e.g., https://searx.be)
  5. API key is optional (only needed if your instance requires authentication)

Public Instances

Public SearXNG instances are often rate-limited or unreliable. Self-hosting is recommended for consistent use.

Configuration

Max Results

Set how many search results to fetch per query: 1–10. Default is device-dependent. More results give the model more context but cost more tokens.

Enable/Disable

Toggle Enable Web Search on the Web Search settings page. When disabled, the model cannot call the web search tool.

When you ask a question that needs current or external information, the model automatically calls web search:

  1. Search: Model calls the search API with a query it formulates
  2. Fetch: Model can optionally fetch full page content from result URLs
  3. Synthesize: Model reads the results and integrates them into its response

You'll see each search and fetch as inline tool cards in the conversation.

Example

You: "What's the latest version of Python?"
Model: [Searches "latest Python version 2026"]
       [Reads result]
       "Python 3.14.0 was released in October 2025..."

Web Fetch

Beyond search, the model can fetch and read specific web pages. When the model encounters a URL in search results, it can call web_fetch to retrieve the full page content:

  • The fetched content is converted to markdown
  • The model processes it and extracts relevant information
  • Fetch results are shown as tool cards

Privacy Considerations

When using web search:

  • Your queries go to the search provider (Brave, Serper, etc.), not to Agora
  • Agora does not log or store your search queries (except in the conversation itself)
  • SearXNG self-hosting gives you the most privacy — queries stay on your infrastructure