Redcast Web search and page reading
Search engines, linked pages, PDFs and what the model can actually read.
Turn on Web Search in AI settings or Quick AI's ⌘K menu. On supported API models, Redcast supplies web_search and page-reading tools. A model must support tool calling to use them; switching search on cannot add that capability to a text-only route.
Search engines
The built-in search path can try Exa, Parallel and DuckDuckGo without a user-supplied key. You can configure your own SearXNG instance, Brave or Tavily connection. Engines are tried in the configured order, allowing another engine to answer when one fails.
No-key access is still a remote request to a search service. It can be rate-limited or unavailable. Configure your own supported engine if you need a particular service.
Ask for sources
A useful question names what you need verified: “Find the official installation instructions for this tool, read the relevant page, and link it in your answer.” Search results and tool rows let you see what the model actually looked up.
A search snippet is not the full page. Redcast gives the model read_page in plain chat and web_fetch in native agents to read sources after searching. Pages become compact Markdown with their links; long pages can be read in parts. PDFs are read by pages.
Pages that cannot be read
A page that depends on JavaScript to populate its content may return only an empty shell. Redcast reports that so the model can try another source. This tool is not a logged-in browser and cannot assume access to your accounts, cookies or pages behind authentication.
Hidden page text is dropped. Fetched content is presented as source material, not as instructions for the agent.
Scope after private context
A native agent that has read personal files, Notes, screen context or attachments uses a narrower URL scope. Unattended runs use it from the start. Fetches are limited to links in the user's request, search results and same-site links from pages already read, with a shared limit of eight pages per turn/run as enforced by the tool context.
A refused fetch can therefore mean the URL is outside the allowed source set, not that your internet is down. Provide the relevant public link explicitly or ask the agent to search for it. Do not embed secrets in a URL.
Troubleshooting
If the answer contains no tool activity, check that Web Search is on and the selected model supports tools. If one provider fails, try another configured search route. If a page is an empty shell, ask for an official static page, documentation export or PDF.