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

# Web search

> How the agent reaches out to the open web — local DuckDuckGo by default, or Rumus's cloud search API for better results.

The agent can search the web and fetch specific URLs. There are two engines under the hood, and a single toggle decides which one is used.

## The two modes

| Mode                       | Network                  | Where it goes                 | Cost                                         |
| -------------------------- | ------------------------ | ----------------------------- | -------------------------------------------- |
| **Local DuckDuckGo Lite**  | Direct from your machine | DuckDuckGo's lite search page | Free                                         |
| **Cloud Rumus search API** | Via Rumus                | A paid search backend         | Tiny per-query, included with built-in usage |

Both return results in the same shape — title, URL, snippet — so the agent doesn't notice the difference. You probably will: cloud results are more reliable and richer, local results are free.

## Toggle the mode

**Settings → AI → General → Cloud Web Search** — toggle it.

When **off**: the agent uses local DuckDuckGo Lite. No login required. Free.

When **on**: the agent uses the Rumus cloud search API. Requires sign-in. Comes with built-in subscription usage tracking.

## Local DuckDuckGo Lite

The default. Rumus fetches the [DuckDuckGo Lite](https://html.duckduckgo.com/) HTML search page directly from your machine and parses the result rows out of the page. Pros and cons:

* ✅ Free. No account needed.
* ✅ No third-party tracking via Rumus.
* ❌ Search quality is what DuckDuckGo Lite gives you — fine for many queries, weaker for niche or technical queries.
* ❌ Subject to whatever rate limits DuckDuckGo applies — heavy use can occasionally trip them.
* ❌ Page format changes upstream can break parsing until Rumus updates.

## Cloud Rumus search API

When enabled, search queries go through Rumus's cloud search API, backed by a commercial search backend. Results include:

* Title, URL, snippet (same as local).
* **Extra snippets** — additional context paragraphs from the page.
* **Age metadata** — how recent the result is.

The cost is a tiny per-query fee that's part of your normal Rumus billing. There's no separate "search bill"; it shows up alongside any built-in model usage.

You need to be **signed in** for cloud search to work — the API uses your account token.

## Web fetch (separate)

Search and fetch are two different tools:

* **Search** — give it a query, get back a list of result URLs.
* **Fetch** — give it a URL, get back the page contents (text-extracted).

The agent typically searches first to find a URL, then fetches the most promising result. You can also direct it: "fetch `https://example.com/...`" skips search entirely.

The fetch tool uses a headless Chromium engine that downloads on first use (around 150–200 MB) — Rumus prompts you to install it the first time the agent wants to render a real web page. Skip it (and lose the ability to fetch JS-rendered pages) by leaving the engine uninstalled.

## When to enable cloud search

If you're a daily AI user and you've noticed the agent missing recent or niche information when searching, flip cloud search on. The improvement is most visible on:

* Recent events / changelogs / version-specific docs.
* Highly technical queries where local search throws up SEO content.
* Searches that benefit from richer snippets (the model can pull better context without an extra fetch).

For light occasional use, local DuckDuckGo Lite is fine.

## Privacy

* **Local mode** — your search query is sent to DuckDuckGo. Standard DDG privacy applies. Rumus doesn't proxy or log it.
* **Cloud mode** — your search query is sent to Rumus's search API. Rumus logs the query for billing / abuse monitoring, but not the resulting page contents.

For the deeper view of how Rumus handles your data, see [Vault & encryption](/docs/account/vault).

## Tips

* **Be specific in your prompt.** "Find recent issues with HTTP/3 in Caddy" beats "tell me about Caddy" — both for the agent and for the search engine, regardless of which mode you use.
* **For one-off niche searches, point at the URL directly.** If you already know the page (a man page, a vendor doc), fetch it instead of searching.
* **For repeated specialized lookups, prefer an MCP tool.** Internal docs, ticket systems, and similar are better served by [an MCP integration](/docs/ai/mcp) than by general web search.

## Next steps

<CardGroup cols={2}>
  <Card title="Agentic execution" icon="robot" href="/docs/ai/agentic">
    Where search and fetch fit in the agent's tool loop.
  </Card>

  <Card title="MCP integration" icon="plug" href="/docs/ai/mcp">
    For internal data, custom tools beat general web search.
  </Card>
</CardGroup>