Web search and browsing
Web search works out of the box: with no configuration at all, the AI searches via DuckDuckGo. Browsing (a real browser the AI can navigate and click) is off by default and enabled per instance.
Web search
Configure in Settings → Admin → Web: a master toggle, plus a provider set to auto or a specific one.
| Provider | Needs |
|---|---|
| DuckDuckGo | Nothing; always available |
| Exa, Perplexity, Tavily, Brave, Firecrawl | API key |
| SearXNG | Base URL of your instance, no key |
| Chat Completions endpoint | Key + URL + model (any search-capable chat API) |
In auto mode, Computer uses the first configured provider in this priority order: Exa → Perplexity → Tavily → Brave → Firecrawl → SearXNG → DuckDuckGo. Keys go in the UI or environment variables (EXA_API_KEY and friends; see environment variables).
Browsing
Off by default; master toggle in the same Settings → Admin → Web page. Three providers:
- local: drives a browser already installed on the machine (Chrome, Chromium, Brave, or Edge) headless over the Chrome DevTools Protocol. The browser is auto-discovered and auto-launched; the CDP URL defaults to
http://localhost:9222. This is not Playwright: no browser download, no heavy dependency, just thewebsocketspackage. - firecrawl: Firecrawl's cloud API (key required).
- browser_use: Browser-Use's cloud API (key required).
With browsing enabled the AI can navigate pages, click elements, fill forms, and take screenshots. Browser sessions idle out after 10 minutes by default.
Browser tabs in the UI have a tab source setting: a managed browser, or your personal Chrome profile with its logins and extensions.
A browser signed in to your accounts can act as you: post, buy, delete. Keep the AI on the managed browser unless a task genuinely needs your logged-in profile, and watch those tasks closely.