Agent ·
Your Terminal's AI Can Google Things Now, and It Has to Show Its Work.
Every TRMNLY chat has a web search toggle as of 0.1.103. Claude chats use Anthropic's own search and list numbered sources, other providers fall back through your search keys, and none of them return a silent "no results".

An agent that can't look anything up is stuck with whatever it learned in training: last year's API, a deprecated flag, a library version that no longer exists. On October 13, TRMNLY 0.1.103 gave every chat a way out.1
One toggle, every chat
Each chat now has a globe toggle, plus a /search command, that turns on web search. Claude Code and Codex chats follow the same toggle.1
Claude cites its sources
Claude chats use Anthropic's own web search tool. The transcript shows a "Searched: …" row followed by a numbered Sources list, so you can see where an answer came from.1 Anthropic's documentation describes the same thing: the web search tool gives Claude current information from the web and returns citations for its sources.2
Every other model gets a fallback chain
Other providers use the app's own search. It tries whichever search key you've added (Brave, Tavily, Exa, Serper or Perplexity), and if you haven't added any, it falls back to DuckDuckGo.1,3
The search either returns results or names the error. It never gives you a silent "no results".1 That matters more than it sounds. An agent that thinks a search came back empty will happily make something up. An agent that's told the search failed knows it doesn't know.
Why this is the right default
An answer that cites its sources is one you can check. An answer that doesn't is one you have to trust. A terminal agent that edits your code should fall in the first group, which is why the sources list is on by default rather than hidden behind a setting.
Sources
- 1.TRMNLY 0.1.103 release notes, "Claude searches the web" (October 13, 2026). trmnly.dev/updates/stable/appcast.xml
- 2.Anthropic, "Web search tool." docs.anthropic.com
- 3.Brave, "Brave Search API." brave.com