Web search

Web search lets Strata pull live information from the public web mid-conversation — news, current prices, documentation, anything published online.

Turning web search on

Web search is a per-conversation toggle in the chat composer:

• Click the plus button (+) on the left of the input and toggle Web search.
• When it is on, the plus button turns blue so you remember web is enabled.
• Your most recent choice is saved as your default for new chats, so you only set it once.

Your organization sets the starting default for web search. If your admin has it on, new chats begin with web search enabled; you can still turn it off for any single conversation from the composer.

When web search is on, every message you send is eligible to trigger a web query. The AI decides whether the question needs one — it does not search for every message.

How Strata uses web search

When the AI determines a web search would help, you see:

• An action step in the trail that reads Searching the web… with the query the AI ran, then Found N results once it returns.
• A panel under the step showing the top results — titles, URLs, and favicons.
• After the search, the AI reads the results and incorporates them into its reply.

Where the AI uses web information in its reply, citation chips appear inline — small numbered badges tied to a source. Hover a chip to see the source title and host; the source link opens in a new tab. Where the AI links a source by name in the text, that link is also clickable and opens in a new tab.

When a reply used the web, a Sources button also appears in the message's action row, showing the favicons of the top sources. Click it to open a side panel listing every source the answer drew on.

Image results

For visual queries Strata may also return a contextual gallery of related images alongside the answer. Each web query automatically fetches related images, and the AI places small image sets inline next to the topic they illustrate. Image results stream into the final answer just like text — they do not appear during the action steps.

When the AI uses web search vs. other sources

Strata draws on multiple sources for each answer:

• Database (SQL Server) data — for live data questions when a connection is active.
• Connected files — uploaded attachments, OneDrive content, Outlook emails, and your organization's Knowledge Base.
• AI's training knowledge — for general explanation, code, definitions, history.
• Live web — for current events, recent prices, today's news, rapidly-evolving information.

With web search off, the AI relies on training knowledge for anything that is not in your data — which means it may be out of date for time-sensitive topics. With web search on, the AI prefers live results for questions where freshness matters.

Note

Web search is a property of the conversation, not the model. It works the same whichever model from your organization's catalog you are using.

Privacy and routing

• Web searches happen on the server, not in your browser.
• Search queries are logged as part of the action-step trail for the conversation. If your account has "Save AI responses" off, the trail is discarded along with the rest of the chat.
• Web search also works in voice mode when it is available on your deployment — the AI can look something up mid-conversation and read you the answer.

Usage accounting

A web search itself is a lookup, not an AI model call, so it does not consume tokens. But the results the AI reads back are part of its tool-use loop, which does count toward the token usage your organization sees in the admin Usage page. There is no separate "web searches" meter.

Tips

• Ask for a date. If you want the freshest source, say "as of this week" or "most recent". The AI biases the search query accordingly.
• Ask for citations explicitly. If you want every claim sourced, say "cite sources for each claim." Strata adds more citation chips.
• Disable for confidential work. If you do not want any prompt-derived query going to a search engine, leave web search off for that chat.

Related

• Chat overview — the chat interface
• Models — the model catalog your organization enables
• Incognito — web search still runs in incognito (but nothing is saved)
• Voice — real-time AI voice mode, which can also run web searches