Skip to main content

Overview

Web search allows your IllumiChat assistant to fetch live information from the internet during a conversation. When enabled, the assistant can search for current events, verify facts, find documentation, and reference up-to-date sources — all without requiring you to upload documents to a knowledge base. IllumiChat uses Exa for web search, which provides semantic search results optimized for AI consumption.

When to Use Web Search vs Knowledge Base

ScenarioWeb SearchKnowledge Base
Current events and newsBestNot suitable
Your internal documentationNot suitableBest
Public product documentationGoodBetter (more reliable)
Competitor researchBestNot suitable
General knowledge questionsGoodNot applicable
Proprietary company dataNot suitableBest
You can enable both web search and a knowledge base on the same assistant. The assistant will decide which tool is most appropriate for each query.

Per-Assistant Configuration

1

Open Assistant Settings

Navigate to your workspace and select the assistant you want to configure. Click Settings.
2

Enable Web Search

In the assistant configuration panel, toggle Web Search to enabled. This allows the assistant to use web search when it determines external information would be helpful.
3

Save Changes

Click Save to apply the configuration. The assistant can now search the web during conversations.

Per-Message Toggle

Even when web search is enabled on an assistant, users can control it per-message. In the chat interface, click the globe icon next to the message input to toggle web search for that specific message. When using the API, set webSearch: true in the message payload:
{
  "content": "What are the latest AI model releases this week?",
  "webSearch": true
}

How Citations Work

When the assistant uses web search, it includes source citations in its response. Citations appear as numbered references linked to the source URLs. Example response:
According to recent reports, the new model achieves 95% accuracy on the benchmark [1]. This represents a significant improvement over previous versions [2]. Sources:
  1. https://example.com/article-about-model
  2. https://example.com/benchmark-results
The assistant automatically formats citations and includes the source URLs so users can verify the information.

Tips for Better Search Results

Write Clear Questions

The assistant translates your question into a search query. More specific questions produce better results.
Less EffectiveMore Effective
”Tell me about pricing""What is Stripe’s current pricing for payment processing?"
"What’s new?""What AI features were announced at Google I/O 2025?"
"How does auth work?""How does Auth0 implement refresh token rotation?”

Guide the Assistant

You can explicitly instruct the assistant to search for specific topics:
  • “Search for the latest Next.js 16 release notes”
  • “Find the official React documentation on Server Components”
  • “Look up current best practices for PostgreSQL indexing”

Combine with Knowledge Base

For the best results, use web search alongside a knowledge base:
  1. Upload your internal documentation to a project (knowledge base)
  2. Enable web search on the same assistant
  3. The assistant uses the knowledge base for company-specific questions and web search for external information
Web search results are not stored in your knowledge base. Each search is performed in real time and the results are only used within that conversation.

Limitations

  • Rate limits — Web search adds latency and is subject to additional rate limits. See the Rate Limits page for details.
  • Dynamic content — Search results reflect publicly indexed web pages. Content behind login walls, paywalls, or JavaScript-rendered pages may not be available.
  • Accuracy — While the assistant cites sources, web content can contain inaccuracies. Users should verify critical information from the source URLs.
  • Recency — Search results are typically a few hours to a few days behind real-time events, depending on how quickly content is indexed.

Next Steps