> ## Documentation Index
> Fetch the complete documentation index at: https://docs.ironlabs.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Web

> Ground replies in live web search results, with citations.

The **Web** source sends your prompt out to a real-time web search (Exa primary, Parallel fallback), reads the top results, and lets the model answer with those pages as context.

Toggle from the `@` menu, or stack with other sources for layered grounding.

<Note>
  **Screenshot coming.** Web source toggled on, citations under reply.
</Note>

## When to use it

* Anything **fresh** — news, releases, product pages.
* **Facts you don't want hallucinated** — prices, version numbers, contact details.
* **Comparisons** where one or both items are recent.

## How it works

1. You toggle Web in `@`.
2. You send a prompt.
3. Foundry runs a web search behind the scenes.
4. The top results are read into context.
5. The model answers, with each claim linked to the page that backs it.

## Good for

* *"What did OpenAI announce yesterday?"*
* *"Compare the latest macbook pro and m4 air specs"*
* *"Who's the current CEO of Mistral?"*

## Limits

* The search is broad — for deep, multi-step research, use [Deep Research](/foundry/sources/deep-research) instead.
* Results are only as fresh as the search index — usually minutes to hours.
* Foundry pulls excerpts, not full pages. If a page needs deeper reading, follow the citation.

## Related

<CardGroup cols={2}>
  <Card title="Deep Research" icon="magnifying-glass-chart" href="/foundry/sources/deep-research">
    Multi-step grounding.
  </Card>

  <Card title="Reddit" icon="reddit" href="/foundry/sources/reddit">
    For lived-experience answers.
  </Card>
</CardGroup>